EV battery degradation: what changes in a pack, and what the federal record actually measures

A used electric car loses range as it ages. Almost everything written about that sentence is either a manufacturer’s reassurance or a number somebody made up, and the reason is not laziness. It is that the federal record, which is where this site takes its figures from, does not measure the thing. There is no government dataset of how much capacity a used pack has left. There is no odometer-style reading you can pull with a VIN. The measurement most buyers assume exists does not.

What does exist is worth more than another invented percentage. The EPA fuel-economy file contains a pair of columns that quantify exactly how misleading an observed range figure can be, and NHTSA’s complaint export contains a precise account of what owners report when they think their battery is dying. Neither is a degradation rate. Both change what a buyer should do.

This page is the mechanism page for the electric articles on this site. It covers what physically changes in a pack, what a driver can and cannot see from the dashboard, what the federal record holds, and where it is empty. Where a number is printed it comes from a dated pull that is archived and re-derivable. Where a number does not exist, that is said rather than filled in.

Range is not capacity, and the gap between them is enormous

Start with the arithmetic, because everything else on this page hangs off it. A traction battery stores energy, measured in kilowatt-hours. A car consumes energy at some rate, measured in kilowatt-hours per hundred miles. Range is the first divided by the second. Degradation attacks the numerator. Driving attacks the denominator. The dashboard shows you the quotient and does not tell you which one moved.

That would be a pedantic distinction if the denominator were stable. It is not, and the EPA file lets us say by how much, because EPA rates the two cycles separately and publishes both. The bulk export behind fueleconomy.gov — 50,242 rows across 84 columns, last modified on 7 August 2026 — holds 1,572 records EPA classifies as electric, spanning model years 1998 to 2027. Of those, 1,386 carry both a city range and a highway range, and those run from model year 2012 to 2027.

Take each of those 1,386 records and measure the distance between its own city range and its own highway range, as a share of its own combined range. The median is 12 per cent. The lower quartile is 4.7 per cent and the upper quartile is 16.8 per cent. The ninetieth percentile is 20.7 per cent, and the single widest record in the file sits at 29.7 per cent. In miles, on cars whose median combined range is 285, the median gap is 31 miles and the upper quartile is 47.

Now connect that to capacity. Range is stored energy divided by consumption, so at a fixed consumption rate a pack that has lost ten per cent of its usable energy delivers ten per cent less range. The relationship runs one to one. Which means the median car in this file has a driving-pattern swing that is arithmetically the same size as losing 12 per cent of its battery, and a car at the upper quartile has one the same size as losing nearly 17 per cent.

Put the two together and the practical consequence is blunt. A buyer who takes delivery of a used electric car, drives it the way the previous owner did not, and watches the range figure fall is watching a number that could have moved that far on a brand-new pack. Of the 1,386 records, 816 have a city-to-highway gap of at least ten per cent of combined range and 526 have one of at least fifteen per cent. That is not noise around a signal. On most of these cars it is larger than the signal anybody is trying to detect in the first few years.

The direction of the gap is the part that catches people out, because it is the reverse of what a lifetime of petrol cars teaches. Of the 1,386 electric records, 1,172 — 84.6 per cent — are rated for more miles in city driving than on the highway. Only 198 go the other way and 16 are level. The same export, restricted to plain petrol vehicles over the same model years, holds 14,426 records, and 14,371 of them — 99.6 per cent — rate the highway figure above the city figure. Seven go the other way. Forty-eight are level.

The reason is not mysterious. A petrol engine wastes fuel idling and wastes it again in stop-start work it is badly suited to, so a steady highway cruise flatters it. An electric motor does not idle, recovers a large part of the energy it puts into a stop, and then meets aerodynamic drag that climbs steeply with speed. EPA measures the consumption side directly: across 1,385 electric records carrying both figures, the median city consumption is 34.9 kilowatt-hours per hundred miles and the median highway consumption is 38, and 1,182 of them — 85.3 per cent — use more energy per mile on the highway than in town.

So the buyer who moves house, changes jobs, or simply starts commuting on an interstate has changed the denominator, permanently, and will see a range figure that never comes back. Nothing has happened to the battery.

The swing is a property of the car, not of the pack

The median hides something useful. The size of the city-to-highway gap varies enormously between manufacturers. Grouping the 1,386 records by make and keeping every make with at least twenty of them gives twenty-one rows, and the spread between the top and the bottom of that list is wider than the whole effect being discussed.

A Kia buyer and a BMW buyer are working with different instruments. On the Kia records the highway penalty is a fifth of the car’s rated range; on the BMW records the two figures are practically the same, and a BMW owner who sees range fall on the motorway is seeing something the label did not predict. Advice written for one of those cars is actively wrong for the other, which is why generic guidance about electric range is worth so little.

Body class explains part of it and not all of it. Small two-wheel-drive SUVs, of which there are 122 records, have a median gap of 17.1 per cent, and standard four-wheel-drive pickups, of which there are 128, sit at 15.7 per cent. At the other end, the 183 records EPA files as Large Cars have a median gap of minus 0.6 per cent — that is, they are rated marginally further on the highway than in town — and the 120 Compact Cars sit at 3.3 per cent. Shape and frontal area move this figure the way anyone would expect.

One honest caution, because it changes how much weight the table can carry. Manufacturers have choices about how the label figures for a given vehicle are derived, and those choices are not uniform across the industry. Some of the make-level spread here is aerodynamics and some of it is methodology. What the table establishes is not that Kia builds a worse highway car than BMW; it is that the published city and highway figures for these two makes stand in very different relationships to each other, and that a buyer comparing an observed range against a remembered label number has to know which relationship applies to the car in front of them. If you want the label figures for a specific vehicle rather than a make-level median, they are on the window sticker, and our page on recovering a window sticker by VIN covers what survives and what does not.

What actually changes inside a pack

None of the above says a battery does not age. It does, in several distinguishable ways, and knowing which one a car has is worth more than any single percentage.

Capacity fade is the one everybody means. The cell chemistry that stores lithium loses some of its ability to hold it. Lithium is consumed forming and re-forming the passivating layer on the negative electrode, some of it becomes electrically unreachable, and the active material itself changes structurally over many cycles. The result is a pack that accepts and returns less energy than it did when new. This is the loss that shows up as range.

Power fade is separate and is often what an owner notices first without naming it. Internal resistance rises, so the pack cannot move charge in or out as quickly. The symptoms are a car that accepts a slower charge than it used to, particularly at a fast charger, and one that limits acceleration or regenerative braking more readily when cold or near the extremes of its charge window. A pack can have healthy capacity and degraded power, or the reverse.

Calendar ageing and cycle ageing are different clocks. A pack degrades because it is used and it degrades because time passes, and the two do not add up in a way that a mileage figure captures. A low-mileage car that has spent six years parked outdoors in a hot climate at a high state of charge has been ageing the whole time. A higher-mileage car that has been cycled gently through the middle of its charge window in a temperate place may be in better condition. This is the single most important reason a used electric car cannot be assessed the way a used petrol car is: on a combustion car, the odometer is a decent proxy for wear on the expensive part. On an electric car it is one of two clocks and often not the dominant one.

Heat is the accelerant, at rest and in use. The chemical processes that consume lithium run faster at high temperature, which is why thermal management is a design decision with a long tail. A pack with active liquid cooling and a pack cooled only by ambient air do not age alike in the same climate, and the difference compounds over years rather than showing up in the first season. For a buyer this argues for knowing where a car has lived, which is a title-history question as much as a battery one.

State of charge matters as much as charging speed. Sitting at a very high state of charge for long periods is harder on cells than sitting in the middle of the range, and repeated deep discharges are harder than shallow ones. Frequent high-power DC charging adds heat and its own stress. None of these is catastrophic in isolation and none is visible in any public record, which is exactly why a seller’s account of how a car was charged is an assertion rather than evidence.

The buffer, and why the number you see is a managed number. Manufacturers do not give the driver the whole pack. Some usable energy is held back at the top and bottom to protect the cells and to give the management system somewhere to go. That buffer means early capacity loss can be partly invisible — the usable window stays the same while the total shrinks into the reserve — and it means the relationship between a pack’s true condition and the figure the car reports is set by the manufacturer’s strategy rather than by physics alone.

The car’s own estimate is an estimate. The management system infers capacity from voltage, current, temperature and history. That inference drifts, and it gets corrected by use — which is why a car that has been sitting on a dealer lot, or has just had its twelve-volt system disconnected, can report a figure that moves after a week of normal driving. A single reading taken on a cold morning after a jump-start is not a measurement of anything. Two readings a fortnight apart, on a car being driven, are worth considerably more.

One more distinction, because it costs people money. The traction battery is not the twelve-volt battery. Electric cars still carry a small conventional battery to run the computers, the contactors and the low-voltage accessories, it still fails the way it always has, and a car that will not wake up is far more often that part than the expensive one. If a car is dead rather than short of range, start with the twelve-volt battery before assuming anything about the pack.

What the complaint file records when an owner names the battery

NHTSA’s Office of Defects Investigation publishes its whole complaint database as a bulk export. The copy behind this page was last modified on 6 September 2026 and carries 2,243,124 rows, of which 2,170,996 concern vehicles rather than tyres, equipment or child seats. Each row is one complaint-component pairing, so a complaint filed under two headings appears twice.

Filtering that to battery-electric passenger vehicles by model name — there is no field that identifies one, which is a point in itself and we will come back to it — gives 42,128 rows, spanning 25,419 distinct complaints across 123 model names. Those rows carry 239 different component headings between them.

Four of those 239 headings name the traction battery. They exist. The taxonomy has the drawer: ELECTRICAL SYSTEM:PROPULSION SYSTEM:TRACTION BATTERY, plus separate headings for its management module, that module’s software, and its thermal-management fan. The earliest complaint filed under one of the four was received in 2018 and the most recent in 2026, so this is not a heading that has just been invented.

Twenty-five rows are in it. Out of 42,128.

That is not because owners are not writing about batteries. Search the narratives for language that names the traction pack specifically — traction battery, high-voltage battery, battery pack, state of health, battery degradation, capacity bars and their neighbours, deliberately excluding the bare word “battery” because it catches the twelve-volt one — and 1,374 rows qualify, covering 786 distinct complaints. Where do they sit?

Five hundred and eighty-three of them are filed under plain ELECTRICAL SYSTEM, a heading that in the same file also carries window switches, wiring, instrument clusters and infotainment. Two hundred and sixty-five are under FUEL/PROPULSION SYSTEM. A hundred and sixty-five are under POWER TRAIN. A hundred and eight are under UNKNOWN OR OTHER, and eighty-nine are under ENGINE — on cars that do not have one.

Two are under a traction-battery heading.

This is not a recent-taxonomy problem that time will fix. Restrict the whole file to rows received in 2024 or later and you get 21,698 of them, of which eight are under a traction-battery heading; of the 896 pack-naming rows in that window, one is. The granular headings are being used elsewhere in the file — 11,767 of the 42,128 rows carry a heading with a colon in it — but 30,361, or 72.1 per cent, are filed under a bare top-level label.

The practical consequence for anyone researching a specific car is direct. If you go to the complaint database, filter to a model, and read what is filed under the battery heading, you will conclude that almost nothing has ever gone wrong. The complaints are there; they are filed somewhere else. Reading the narratives, not the headings, is the only method that works on this subject.

Why none of that is a degradation rate

It is tempting to take 1,374 rows, or 786 complaints, and turn them into a percentage of something. It cannot be done, and the reasons are worth stating precisely because they apply to every complaint-based claim anybody makes about any car.

There is no denominator. NHTSA publishes complaints; it does not publish how many of each vehicle are registered and driving. Without that, a count of filings is a count of filings. A model with more complaints may have more problems or may simply have sold in greater numbers, be newer, be owned by people more inclined to file, or have been the subject of news coverage that prompted filing.

The filings are self-selected in both directions. Nobody files a complaint to report that their battery is fine. People file when something has gone wrong, when it has gone wrong expensively, and often when a manufacturer has already refused them. A file built that way is a description of grievance, not a sample of the fleet.

And the file does not contain the measurement anyway. Of the 1,374 rows that name the pack, 360 — 26.2 per cent — contain any number at all that could be a measurement: a percentage, a kilowatt-hour figure, a count of capacity bars. Narrow that to an explicit state-of-health reading or bar count and it is twenty rows. Twenty, out of 42,128. What owners overwhelmingly write down is an event: the car would not charge, the range collapsed suddenly, a warning appeared, the car was towed. Across all battery-electric rows in the file, 3,048 record that the vehicle was towed and 376 record a fire.

Even the odometer is mostly missing. The export has a MILES field — the vehicle’s mileage at failure — and it is populated on 7,366 of the 42,128 rows. On the other 34,762 it is zero, which in this file means not reported: NHTSA’s own change log records that previously blank numeric fields now show as zero. Where it is reported, the median is 19,507 miles, and among the pack-naming rows that carry it — 184 of them — the median is 21,991 with a quartile range from 7,200 to 65,000. Those are the mileages at which people chose to write in. They are not the mileages at which batteries fail, and a hundred and eighty-four self-selected readings would not establish that even if the field were complete.

One more absence, and it is a strange one. The export carries a FUEL_TYPE field. Its published code list has five values: bifuel, CNG, diesel, gas and hybrid electric. There is no code for a battery-electric vehicle, and across all 2,243,124 rows none is used — the values that actually appear are blank, GS, DS, BF, HE and CN. The federal complaint file cannot tell you that an electric car is electric. That is why the selection for this page had to be made on model names, and it is why any count anybody quotes from this file about electric vehicles rests on somebody’s hand-built list rather than on a field.

The one place a capacity number was going to appear

The emissions warranty is the only warranty that arrives with a used car by operation of federal law, and it now reaches the traction battery. Our page on checking warranty coverage by VIN sets out the periods, who they run to and how a claim is raised; there is no point repeating them here. What belongs on this page is a narrower question: does the rule anywhere require somebody to tell a buyer how much capacity a pack has, or has left?

The answer, read out of the current text rather than out of a search index, is nearly. 40 CFR part 85 subpart V, at the issue of title 40 dated 31 August 2026, contains exactly one requirement that mentions a percentage of usable battery energy. It is in 85.2109, the section governing what a manufacturer must put in the owner’s manual or warranty booklet, and it says that for battery-electric and plug-in hybrid vehicles the explanation must include the manufacturer-defined value for percentage usable battery energy specified in 85.2103(d)(3).

In the same current text, 85.2103(d) runs to two subparagraphs, (1) and (2). There is no (d)(3). The one cross-reference in the subpart points at a paragraph that is not there.

The same reading finds nothing elsewhere. Across the whole of part 85 subpart V the phrase “traction battery” appears zero times. Across the whole of 40 CFR part 86 subpart S — the light-duty emission standards and certification subpart, some 699,470 characters of it — the phrases “battery durability”, “usable battery energy”, “state of certified energy” and “monitor family” each appear zero times. The same four phrases appear zero times in 40 CFR part 1037 subpart B, the heavy-duty equivalent. There is, in the current rules, no minimum retention standard and no requirement that a vehicle report what is left of its pack.

A caution about method, since it is the reason this section says what it says. eCFR’s full-text search still returns a section number for “battery durability” that the current structure of title 40 does not contain. Searching a regulation and reading a regulation are different activities, and a page built on the first would have reported a standard with no current text behind it.

One live provision is worth a used buyer’s attention, because it runs the other way. The subpart allows a manufacturer’s written instructions to identify behaviour expected to unreasonably or artificially shorten battery durability — exceeding the towing capacity is the example given — and says evidence of compliance is generally limited to onboard data logging, though an owner may also be asked to make a statement. Read plainly, that contemplates a manufacturer looking at a car’s own logged history when deciding a claim. On a used car, that history belongs to a previous owner and is not something a second buyer can inspect, correct or account for.

What a buyer can actually observe

Given all of that, here is what is left, and it is more than nothing.

Ask the car for energy, not for miles. Nearly every electric car will display consumption — kilowatt-hours per hundred miles, or its reciprocal — over a trip or a rolling average. That figure is the denominator, and it is directly comparable to the EPA consumption rating for that model, which the file publishes alongside the range. If the car’s observed consumption is well above the rating, the range shortfall is being caused by how it is driven or the conditions it is in, and a capacity conclusion is not available from that data.

Measure a charge, not a guess. The strongest thing a private buyer can do without special equipment is watch a charging session: note the indicated state of charge before and after, and the energy delivered. Energy in, divided by the fraction of the pack it filled, is an estimate of total usable capacity. It is rough — charging losses, temperature and the car’s own rounding all intrude — but it is a measurement of the numerator rather than of the quotient, which is more than the range display gives you.

Treat the range readout as the weakest evidence available. It is a projection built from recent driving. Drive a car gently for ten miles and it will predict a longer range than the same car after ten miles of motorway. A seller who has done a careful low-speed loop before you arrive has moved that number without doing anything dishonest to the car.

Compare against itself, not against the brochure. The comparison that means something is the same car, the same route, the same weather, at two points in time — or the car against another example of the same model year driven the same way on the same day. A comparison against a remembered EPA figure is a comparison against a laboratory cycle, and the table above shows how far the city and highway ends of that cycle can be from each other.

Ask for a battery health report and read what it actually says. Franchised dealers for most electric models can run a diagnostic that reports the management system’s own view of pack condition, and some manufacturers make a version of it available to owners. It is the manufacturer’s estimate rather than an independent measurement, and it is the estimate that will govern any warranty conversation, which is precisely why it is the one to get in writing before money changes hands.

Watch charging behaviour, not just capacity. Power fade shows up as a session that tapers early, a car that will not take its rated peak from a fast charger it should manage, or persistent thermal limiting. On a test drive, a car that noticeably restricts regenerative braking when it is not cold is telling you something. Our guide to arranging a pre-purchase inspection covers finding somebody who works on electric vehicles rather than a general workshop — on this subject the difference is diagnostic access, not enthusiasm.

The order to do it in

Establish what the car is before establishing what its battery is. Model year, trim and pack size determine which EPA figures apply, and a mismatch between the advert and the vehicle ends the conversation cheaply.

Run the free federal checks. Open recalls on electric vehicles frequently concern the battery, its management software or its charging system, and a campaign that has not been completed is a live liability rather than a footnote.

Read the complaint file for that model and model year — by narrative, not by component heading, for the reason established above. You are looking for a pattern of sudden capacity loss, refusal to charge, or towing events, not for a percentage.

Then measure. Consumption against the rating, a charging session for capacity, a dealer health report in writing, and a road test long enough to see how the car behaves when the pack is warm and when it is not.

Do all four and you will know more about the battery in front of you than any dataset in the United States could have told you. Do none of them and you are buying the largest single component in the car on the strength of a number the dashboard made up from last week’s driving.

Common questions

Is there a federal database of electric battery health?

No. EPA publishes range and consumption ratings for new vehicles and does not retest a used one. NHTSA’s complaint export carries 51 fields and none of them records capacity or state of health. The current text of the emission warranty rules contains one reference to a percentage of usable battery energy, and it points at a paragraph that is not in the rule. Nothing keyed to a VIN returns a capacity figure, and any service that claims otherwise is reporting the manufacturer’s own estimate or an inference, not a federal record.

My range dropped a lot this year. Is the battery failing?

Possibly, and the first thing to rule out is the denominator. Across the 1,386 EPA electric records that carry both figures, the median car is rated 12 per cent further in town than on the highway, and 168 of them differ by at least 20 per cent. A change in route, speed, load, temperature or tyre pressure moves observed range without touching capacity. Compare consumption against the car’s own EPA rating before drawing a conclusion, and compare the car against itself on the same route rather than against a remembered brochure number.

Why does my electric car do better in traffic than on the motorway?

Because it does not idle and it recovers energy when it slows down, while aerodynamic drag rises steeply with speed. EPA measures this directly: across 1,385 electric records the median city consumption is 34.9 kilowatt-hours per hundred miles against 38 on the highway, and 1,182 of them use more energy per mile at highway speed. The pattern is the reverse of petrol, where 14,371 of 14,426 records over the same model years are rated better on the highway than in town.

Does high mileage mean a worn-out battery?

Less reliably than on a petrol car. Cells age with use and with time, and the calendar clock does not stop when a car is parked. A low-mileage car kept hot and full has been ageing throughout; a higher-mileage car cycled gently in a temperate climate may be in better condition. Mileage is one input and it is not the dominant one, which is why a pack measurement matters more here than the odometer does.

Can I look up a car’s battery complaints under a battery heading?

You can, and it will mislead you. NHTSA has four component headings naming the traction battery, and across every battery-electric row in the export 25 rows are filed under one of them. Of the 1,374 rows whose narrative names the pack, two are filed under a traction-battery heading and 583 are filed under plain ELECTRICAL SYSTEM. Search the narratives, not the headings.

What is the difference between capacity fade and power fade?

Capacity fade is a loss of stored energy and shows up as range. Power fade is a rise in internal resistance and shows up as slower charging, reduced acceleration and more aggressive limiting of regenerative braking, particularly when cold. A pack can have one without the other, and a range-only assessment will miss the second entirely — which matters, because slow charging is the failure that makes a car annoying to live with long before its range does.

Does the federal emissions warranty guarantee a minimum capacity?

Not in the current rules. Coverage for the battery is coverage against defect and against a nonconformity that would cause a failure of the emission standards — the terms are set out on our warranty-check page — and it is a real thing that transfers with the car. But across part 85 subpart V, part 86 subpart S and part 1037 subpart B, read at the 31 August 2026 issue of title 40, the phrases “battery durability”, “usable battery energy” and “state of certified energy” appear a combined total of two times, both in the warranty subpart, and neither of them sets a retention standard. A manufacturer may offer one in its own warranty; that is a commercial promise, and it should be read in the document that makes it.

How much capacity should a used electric car have lost?

This page will not give you a number, and the reason is the whole argument above. No federal source measures capacity in service, so any percentage-per-year figure is either a manufacturer’s claim, a survey of self-selected owners, or an estimate somebody made and everybody repeated. What can be said with evidence is that the driving-pattern swing built into the EPA ratings has a median of 12 per cent of combined range and reaches 29.7 per cent on one record in the file, which is larger than the early-life loss most people are trying to detect. Measure the car in front of you rather than reasoning from an average.