Symptoms of a Bad Head Gasket: Most Owners Report None

The list of head gasket symptoms is one of the most copied passages on the internet. White smoke from the exhaust, milky oil under the filler cap, bubbles in the expansion tank, a temperature gauge climbing on a full system. It is not wrong. It is the failure described from the workshop side, by people who see the joint after it has been taken apart, and it has been repeated for so long that nobody checks it against what happens to the person driving the car.

So we checked. The National Highway Traffic Safety Administration keeps every complaint an owner files, and it keeps the free text they wrote — not a category, not a checkbox, the actual sentences. That archive is public. We pulled 95,913 of those records across 165 vehicles, found the 1,189 whose narrative contains the phrase “head gasket”, and narrowed to the 723 where the failure is the subject of the complaint rather than something mentioned in passing.

What comes back does not look much like the list. The most striking thing in the corpus is what is missing from it, and the second most striking thing is that the two symptoms everyone leads with are near the bottom of what people actually write down. This page is about that gap: what this failure looks like from the driver’s seat, what it is repeatedly mistaken for, and what to do with a suspicion once you have one.

What the owner record actually holds

A word about where this comes from, because the provenance is the only thing that makes it worth more than the copied list.

NHTSA’s complaints endpoint returns every report filed against a given make, model and model year, and each record carries the owner’s narrative alongside the component category a clerk sorted it into. The category is close to useless for this subject — “engine and engine cooling” holds thermostats, radiators, water pumps and stalling complaints in the same bucket as a failed cylinder head joint, and nothing in the field separates them. The text is the only place the distinction survives.

We pulled 165 vehicles, chosen in two blocks. One block is model years with a public reputation for this specific failure; the other is high-volume ordinary vehicles pulled as a contrast. That selection is editorial and it matters enormously to how the results may be read, which is set out further down. It produced 64,344 complaints in the reputation block and 31,569 in the control block.

Of those 95,913 records, 1,189 mention a head gasket somewhere. But a mention is not a subject: an owner complaining about an airbag who notes a gasket repaired three years earlier is in that count. So the working corpus is narrower — the 723 reports where the phrase appears in the first two sentences, on the reasoning that a complaint about a head gasket names it immediately and a passing reference arrives later. That rule is crude and it is published rather than hidden; it discards 466 records, some of which were genuinely about the failure.

Now the finding. Across those 723 reports, we scanned for eleven families of symptom, each written generously to catch the way owners phrase things rather than the way a workshop would — overheating in all its forms, coolant going missing, rough running, warning lights, stalling, power loss, smoke and steam, smells and visible leaks, oil and coolant mixing, a system that pressurises, a heater gone cold.

402 of the 723 matched none of them. Not one word, anywhere in the narrative, describing anything the owner noticed. 55.6 per cent of complaints about this failure contain no account of how it announced itself.

Some of that is the records rather than the world, and the honest version of this finding says so. These narratives are short — the middle one runs 281 characters, a quarter of them are under 139, and 255 of the 723 are under 200 characters, many of them a single truncated line from the oldest intake records in the archive. Split the corpus by length and the effect is plain: among reports under 200 characters, 187 of 255 name no symptom; between 200 and 499, it is 150 of 276; among the 192 longest reports, it falls to 65. The pattern is real, and it means the 55.6 per cent figure is partly an artefact of how much room people had.

But it does not go away. Even in the fullest third of the corpus — the narratives running past 500 characters, where owners had all the space they wanted and were plainly using it — a third of them still describe no symptom at all. That is the part that is not about record length. It is about what this failure is.

Found, not felt

Sort the 723 reports by how the failure reached the owner and the shape of the thing becomes clearer. Four routes, and every report falls into exactly one.

339 reports — 46.9 per cent, the largest group by a distance — are a bare assertion. A head gasket has failed. No symptom described, and no mechanic, dealer, shop or inspection credited with finding it. The owner writes the conclusion and stops.

208 reports, 28.8 per cent, describe a symptom and name the failure themselves without attributing it to anybody. 113, 15.6 per cent, describe a symptom and then attribute the diagnosis to somebody who looked at the car. And 63, 8.7 per cent, are the purest form of the thing this page is about: no symptom described at all, and the failure explicitly attributed to a workshop. Somebody had the car up for another reason and came back with news.

Put the two attributed routes together and 176 reports, 24.3 per cent, credit a third party with naming the fault. That is a floor rather than a measure — it counts wording, and plenty of the bare assertions were certainly also somebody else’s diagnosis, written down without the attribution. But even as a floor it says something the symptom list does not: for a substantial share of owners, the first they knew of this was a sentence spoken to them.

This is the difference between a head gasket and almost every other thing that goes wrong with a car. A failing wheel bearing makes a noise you can hear from the driver’s seat. A dying battery leaves you standing in a car park. A head gasket, in its early and middle life, does nothing you would notice unless you were looking for it — it moves a small amount of gas into a sealed circuit and a small amount of coolant out of one, and both of those are invisible from behind the wheel until they are not small any more.

Which reframes the question the page title asks. “What are the symptoms” assumes the failure declares itself. Frequently it does not. It gets found — at an oil change, at a pre-purchase inspection, at a service visit for something unrelated — and the useful skill is knowing which quiet, easily-explained-away things are worth taking seriously, not memorising a list of dramatic ones.

The symptoms owners do describe, and the famous ones they do not

321 of the 723 reports name at least one symptom. Here is what they name, counted across the whole corpus.

A warning light comes first, in 91 reports — 12.6 per cent. Misfire or rough running is next at 90, then overheating at 89, then coolant disappearing at 76. Those four are effectively level; the gaps between them are smaller than the noise in a sample this size and nobody should read a ranking into the order. Power loss or limp mode follows at 56, stalling or refusing to start at 52.

Then the drop. A burning smell or a visible leak appears in 20 reports. White smoke or steam — the single image most associated with this failure, the one that leads every list — appears in 14. Oil and coolant mixing, the milky film under the filler cap, appears in 10. A cold heater appears in 7. A cooling system visibly pressurising, bubbling in the expansion tank or pushing coolant out of the overflow, appears in 5.

The five classic tells, added together, appear in fewer reports than warning lights alone.

Most owners who name anything name only one thing: 193 reports name a single symptom family, 83 name two, 32 name three, 11 name four and 2 name five. A car doing all six things at once, which is how the failure gets described in most guides, is not what the record looks like. The record looks like one modest complaint, or nothing.

The check that this survives

Two nameplates dominate this corpus and it would be dishonest not to say so up front. A 1996 Ford Windstar supplies 246 of the 723 reports, 34 per cent on its own. A 2018 Honda Accord supplies 217, another 30 per cent. Between them, 64 per cent. 73 distinct vehicles contribute at least one report and 92 of the 165 we pulled contribute none at all.

So any headline computed across the whole corpus risks being a fact about two cars wearing the costume of a general finding. The rollup was therefore recomputed with both removed, leaving 260 reports from everything else.

The absence holds. 148 of those 260 — 56.9 per cent — still name no symptom, marginally higher than the 55.6 per cent across the whole set. The single largest result on this page is not an artefact of two nameplates.

The ranking does move, and it moves in a way worth knowing. Overheating goes to the top at 37 reports, 14.2 per cent. Coolant disappearing follows at 27. A warning light drops to 24, stalling to 21, and misfire or rough running collapses from 12.4 per cent of the full corpus to 5.8 per cent of the remainder — 15 reports. That is a turbocharged modern car’s vocabulary leaving the sample: limp mode, warning lights and rough running are how a failure presents on an engine with a management system watching it, and they were not evenly spread across the corpus.

The classic tells stay rare in both cuts, which is the point that matters. White smoke or steam is 6 of the 260, 2.3 per cent. Oil and coolant mixing is 7, 2.7 per cent. On a smaller sample they tick up very slightly and remain, in both readings, unusual. If you are waiting for white smoke before you take a suspicion seriously, you are waiting for something that most owners in this archive never reported seeing.

What each sign is also caused by

Every symptom in that list is produced by at least two cheaper things, and that is the practical reason this failure gets found late. Nothing a head gasket does is unique to a head gasket. What follows is not a diagnostic procedure — that belongs to somebody with the car on a ramp. It is the sorting you can do yourself before you decide whether this is worth paying somebody to investigate.

Read down the middle column and the pattern is hard to miss: every alternative is a smaller job. That asymmetry is why the failure is so often left alone. Each individual sign has an innocent explanation available, each innocent explanation is cheap, and a person choosing between explanations without evidence will choose the cheap one almost every time. There is nothing foolish about that. It is only a problem because this particular fault gets more expensive the longer it runs.

Which one they reach for first

Where a narrative names several symptoms, the one it names first is worth counting separately — not because it happened first, but because it is what the owner reached for when they sat down to explain what had gone wrong. Of the 321 reports naming any symptom, overheating leads the telling in 61 of them, misfire or rough running in 56, and a warning light and coolant disappearing in 52 each. Stalling or refusing to start opens 35 narratives, power loss 27.

Compare that against the named-anywhere counts and one family stands out. Overheating is named in 89 reports and opens 61 of them: when it is present, it is almost always the headline. A warning light is named in 91 and opens 52, so it is more often mentioned as context than as the story. Neither of those is a fact about the car. They are facts about which detail an owner considered the important one, and they are worth exactly that much.

It also has to be said plainly that the order in a narrative is the order of the telling and not the order of events. People write the worst thing first. A gauge that went into the red is more memorable than a coolant bottle that needed topping up three times over a summer, so it gets written first even where the bottle came months earlier. Almost none of these records were taken down over the phone by a call handler — only one of the 723 uses NHTSA’s intake template — so what we are reading is genuinely owners typing, with all the compression that involves.

What else is in the same sentence

The other words in these narratives are as revealing as the symptom vocabulary.

The most common thing an owner brings up alongside the failure is a recall: 139 of the 723 reports, 19.2 per cent, mention one — asking why there has not been a campaign, referring to a campaign on a neighbouring model year, or arguing that there should be. That is a large share, and it is people using the complaint form for the purpose it exists for, which is to tell a regulator that something looks like a pattern.

They are mostly going to be disappointed, and the severity flags in the same corpus explain why. Across all 723 reports there were 4 crashes, 3 fires, 2 injuries and no deaths. A safety recall addresses an unreasonable risk to safety. A joint that starts to weep, expensive as the consequences are, does not generally produce one — it produces a bill. That is not indifference on anybody’s part; it is the boundary of what the recall system is for. If you want to know whether free work is sitting on a specific car, a VIN check against the federal recall database takes about two minutes and costs nothing.

After recalls, the co-mentions are mostly about consequences and neighbours. An engine replaced or rebuilt appears in 51 reports, 7.1 per cent. A turbocharger appears in 49, almost all of them modern cars. A tow appears in 21, a radiator in 18, a water pump and a timing belt or chain in 10 each, a thermostat in 9, a catalytic converter in 6. Warranty having run out appears in 7, and the car being scrapped, sold on or written off appears in 2 — a small number that mostly reflects the fact that people file these reports while they are still deciding, not afterwards.

131 reports, 18.1 per cent, contain a repair figure the owner typed in. We are not going to print any of them, and that is a settled policy on this site rather than a judgement about these particular numbers: a price set in type keeps looking authoritative long after it has stopped being true, and one owner’s quotation says nothing about the next person’s. What the count does establish is that money is close to the surface of this subject. Nearly one report in five reaches for a figure unprompted, in a form that never asks for one.

Where these reports end up, and why searching by component misses them

If you go looking for this failure in the complaints database yourself, the component filter will lose a good share of it.

Of the 723 reports, 398 carry the tag “engine and engine cooling” and 299 carry “engine”. Those are two different buckets for what is, in every one of these narratives, the same part. Beyond them the reports scatter: 55 are tagged fuel or propulsion system, 51 power train, 40 “unknown or other”, 27 electrical system, 15 service brakes, 12 hydraulic.

Only 328 of the 723 are filed under “engine and engine cooling” and nothing else. 171 carry more than one component tag. And 27 carry no engine-related tag whatsoever — a narrative plainly about a failed head gasket, filed under something else entirely, invisible to anybody browsing the obvious category.

This is worth knowing for a practical reason. When you are researching a specific model year before buying, filtering the complaint database by component is the natural move and it will systematically undercount exactly this failure. Reading the narratives is slower and it is the only method that works. It is also why we do not treat component tallies as evidence about head gaskets anywhere on this site.

One more filing detail, offered because it colours everything above. The median gap between the date of the incident and the date the complaint was filed is 17 days. 236 of the 680 dated reports — 34.7 per cent — were filed within a week, and 46 were filed more than a year later. People write these while it is fresh, which is good for accuracy and bad for outcomes: most of these narratives end before the story does.

Three things this data will not tell you, and we are not going to guess

The temptation with a dataset this size is to keep squeezing. Three questions readers reasonably want answered cannot be answered from it, and saying so is more useful than an estimate dressed up as a finding.

At what mileage does this happen. The only odometer reading in a complaint record that can be trusted is the one written to a fixed intake sentence by a call handler, because everything else is an owner’s recollection in prose — sometimes the reading at failure, sometimes when the trouble started, sometimes the interval between two failures. Exactly 3 of the 723 narratives carry that sentence. There is no mileage figure to report here, and rather than reach for a looser pattern and manufacture one, the answer is that this pull cannot say.

Whether it is a summer failure. It is intuitive that a cooling-system fault would cluster in hot months. It does not, in these records. Of the 687 reports carrying an incident date, 174 fall in winter, 196 in spring, 160 in summer and 157 in autumn — a spring bulge if anything, and summer is the second-lowest quarter. That is close enough to flat that the honest reading is no seasonal signature at all. Whatever brings owners to file, warm weather is not obviously it.

Which cars this happens to. This is the important one. The vehicles in this pull were chosen by us, on the basis of reputation, specifically to produce a corpus large enough to read. That is a purposive sample and it makes every cross-vehicle comparison in it meaningless as evidence. The demonstration is in our own numbers: the reputation block returned 7.3 subject reports per thousand complaints and the control block returned 8.1. The block we picked for its head gasket reputation produced a slightly lower rate than the block we picked as a contrast. Nothing follows from that about either group of cars — it follows from the fact that a purposive sample cannot measure prevalence, which is precisely why we are not publishing a list of worst offenders. The correct source for how a specific model year behaves is that model year’s own complaint record, read narrative by narrative.

If you think this is your car

The evidence above points at one practical conclusion, and it is not a list of things to check under the bonnet.

Take the quiet version seriously. The single most common presentation in this archive is no presentation at all, and the second is one modest symptom on its own. If your car has been drinking coolant over a few months with nothing on the drive, that is not a small anomaly to keep topping up. It is, in these records, one of the two or three most common ways this failure actually shows itself.

Do not wait for confirmation from the famous symptoms. White smoke and milky oil arrive late in this failure’s life, when coolant is moving in real volume or has been getting into the oil for a while. They appear in a small minority of these reports, and by the time they appear the question has usually changed from whether to repair the joint to whether the engine is still worth repairing. Waiting for them is waiting for the expensive version.

Get the suspicion tested rather than argued about. There is a specific test for combustion gas in the coolant that answers this question directly, and it is quick and cheap next to what follows it. Our page on what drives a head gasket bill explains the four tests and what each of them can and cannot rule out. The point of naming it here is only that a suspicion is worth converting into an answer early, because the two wrong answers cost very differently: authorising major work on a car that needed a thermostat is bad, and driving a car that is slowly filling its oil with coolant is worse.

If the gauge is climbing right now, stop reading and deal with that first. An engine that is actually overheating is a situation with its own rules, and the first two minutes largely determine what you are left with; what to do when the temperature gauge climbs covers it properly, including why the cap is the classic way people get badly hurt.

And keep the receipt trail. A car with a documented gasket repair is worth considerably more to the next buyer than one where the work was done and nobody can prove it, because a buyer with no paperwork has to assume the worst version of the story. That is worth remembering while you are deciding, not afterwards.

If you are looking at one to buy

Everything above changes shape when you are the one considering the purchase, because the failure’s habit of hiding is now working against you rather than the seller.

Arrive before the engine is warm. This is the single most useful thing a buyer can do about this subject and it costs nothing but a phone call. A pre-warmed engine has already burned off the condensation, cleared the cylinder that fouls overnight, brought the gauge to its normal place and stopped whatever it does in its first minute. A seller who insists on having the car running when you arrive may simply be being helpful. You still want the cold start, and it is reasonable to ask for it.

Look for the arithmetic of neglect rather than for symptoms. Coolant that is the wrong colour or well below the mark, a residue crust around a hose end, a filler cap that has clearly been off recently on a car with no service history, a temperature needle that sits somewhere unusual. None of those proves anything on its own; all of them are questions to ask.

Treat “it just needs a head gasket” as a description of a decision, not a defect. Cars advertised that way are common because the fault reliably arrives at a point in a car’s life when the sums have stopped being obvious, and the seller has usually already done the arithmetic and reached a conclusion. That does not make it a bad buy — it makes it a buy where you are taking on an unknown that cannot be priced until the head is off.

Remember what an as-is sale means. A used car sold as-is under the FTC’s Used Car Rule carries no dealer obligation for what goes wrong afterwards, and a private sale generally carries even less. That is exactly why the inspection has to happen before the money moves; our guide to what a pre-purchase inspection should cover sets out what to insist on. A cooling system pressure test and a combustion gas check are cheap to add and they are the two that bear directly on this.

Finally, the counterintuitive one. A car with documented head gasket work already done is often a better proposition than an identical car with no such history, provided the paperwork is real and shows what else was done at the same time. The job reaches most of the front of the engine, so a proper repair tends to renew several things at once. The version to be wary of is not the repaired car. It is the car where a bottle of sealant went in and nobody wrote anything down.

Common questions

What is the first sign of a blown head gasket?

In this archive, most often there is no sign at all — 402 of 723 reports about the failure describe nothing the owner noticed. Where owners do describe something and we can see which they mention first, overheating opens the account most often, in 61 of the 321 reports naming any symptom, with misfire or rough running, a warning light and coolant disappearing close behind. But the order in a written complaint is the order of the telling, not the order of events, so treat that as what people considered worth leading with rather than as a chronology.

Can a head gasket fail without the engine overheating?

On this evidence, routinely. Overheating is named in 89 of the 723 reports, which leaves the substantial majority of them not naming it. Some of that is short records, but it also reflects how the failure works: a joint can pass coolant into a cylinder, or combustion gas into the coolant, in quantities too small to move a temperature gauge for a long time. A normal-looking gauge is not a clean bill of health for this.

Does white smoke from the exhaust always mean a head gasket?

No, and the more useful finding here runs the other way: a head gasket failure very often does not mean white smoke. It is named in 14 of the 723 reports, 1.9 per cent, and in 6 of the 260 once the two dominant nameplates are removed. Meanwhile every engine produces visible white vapour on a cold morning, which is condensation and entirely normal. The distinguishing question is whether it is still there once the exhaust has warmed through.

Is milky oil under the filler cap proof of a head gasket failure?

It is not, and it is one of the most over-read signs in the whole subject. A creamy film on the underside of a filler cap is condensation, and it is very common on a car that does short journeys in cold weather. Oil and coolant mixing is named in just 10 of the 723 reports. What changes the picture is emulsion visible through the oil itself or an oil level that has gone up rather than down, and that is a much more serious observation than a filmed cap.

How long can you drive with a failed head gasket?

There is no answer to that in miles and anyone offering one is guessing. What the failure does is progressive rather than sudden: gas going into the coolant creates hot spots that make the leak worse, and coolant getting into the oil attacks bearing surfaces that no amount of work on the cylinder head will reach. That is the reason the question matters. The repair on a joint that has just started to pass and the repair on a joint that has been passing for months are two different jobs, and the second one is often not worth doing.

Will a scan tool find a blown head gasket?

Not directly. There is no fault code for a failed head gasket. A scan tool reports what the engine management system noticed — a misfire on one cylinder, a coolant temperature reading that does not behave, a lean condition — and those are consequences that have several other causes. It is worth reading the codes because they narrow the field, and it is worth knowing that a clean scan settles nothing here. A warning light was named in 91 of these reports, more than any other symptom, but a light tells you the car noticed something, not what.

Would this show up on a vehicle history report?

Only if somebody reported it. A history report reflects records that reached it — title events, insurance claims, and service entries from workshops that share data. A head gasket repaired at an independent garage and paid for in cash generates no record anywhere. This is one of the more significant blind spots in any history report, and it is the argument for an inspection rather than a substitute for one.

Are head gaskets ever recalled?

Very rarely, and the severity data in this corpus shows why. Across 723 reports there were 4 crashes, 3 fires, 2 injuries and no deaths. A recall campaign addresses an unreasonable risk to safety, and a gasket that begins to weep is a durability problem with a large bill attached rather than a safety defect. Owners know this and push back at it — 139 of the reports raise a recall directly. What does turn up in recall records are the components that can cook a cylinder head, like coolant pumps and valves, which is worth a VIN check on any car with cooling history.

Should I buy a car that has already had the head gasket replaced?

Often yes, if the work is documented. The job requires taking off most of the front of the engine, so a proper repair usually renews several other things at the same time, and a car with real paperwork for it can be a better buy than an identical one with an unknown history. The version to avoid is the car where the symptom was quietened with a sealant and nothing was written down — and the way to tell the difference is the receipts, not the engine bay.