Symptoms of a Bad Spark Plug: Which Ones Actually Mean Anything

Search this question and you will get the same page eight times: engine misfires, rough idle, trouble starting, poor acceleration, high fuel consumption, check engine light. The list is not wrong, exactly. It is unfalsifiable. Every one of those things is also what a failing coil does, and a clogged injector, and a vacuum leak, and a tired fuel pump, and a stretched timing chain, and half a dozen other faults that cost a great deal more to put right. A list that describes every engine problem equally well tells you nothing about the one in front of you.

What would be useful is a way of ranking those symptoms by how much each one actually narrows the field. That requires a comparison, and a comparison requires data that nobody writing a symptom list has bothered to go and get.

It exists. Every year, tens of thousands of American drivers file a complaint with the National Highway Traffic Safety Administration, and a large share of those complaints are free text: an owner, describing what their car did, in the order they noticed it, with no template forcing the vocabulary. That archive is public and it is queryable by make, model and model year. It is the closest thing there is to a large sample of people describing engine faults in their own words.

So this page is not another list. It is a measurement of which items on the list are worth anything, made against a control group, with the sample sizes printed and the things it cannot establish stated as plainly as the things it can.

What we counted, and what we counted it against

The pull ran in September 2026 against NHTSA’s complaints endpoint, one request per make, model and model year, with every response archived unmodified alongside its request URL and retrieval date. The vehicle list is 16 models across 11 makes — full-size pickups, mid-size saloons, compact crossovers and hatchbacks — chosen on two criteria and no others: they sell in large numbers in the United States, and they are sold overwhelmingly with spark-ignition petrol engines. A diesel has no spark plugs and would have polluted the sample. Four model years, evenly spaced from 2012 to 2018, so the sample contains cars that have had time to reach a plug interval as well as cars that have not.

That returned 42,679 distinct complaints. 11,028 of them — 25.8 per cent — carry a component tag in the engine family. Within those, 1,082 narratives name ignition hardware or a misfire: a spark plug, an ignition coil, a coil pack, or the word misfire in any of its spellings. That is 9.8 per cent of the engine file. Across the whole pull, including complaints tagged to other systems, 1,201 narratives mention one of those things.

Inside that group of 1,082, the vocabulary splits in a way worth noting before anything else. 849 narratives name a misfire, which is 71.7 per cent of the group. 557 name a spark plug. Only 142 name an ignition coil or coil pack, 11.6 per cent. Owners and the shops they quote reach for the symptom word far more readily than for either piece of hardware, and when they do name hardware, the plug is named about four times as often as the coil that fires it. Whether that reflects what was actually replaced or simply which part people have heard of, the file cannot say.

The control group is the whole point

Here is the part that separates this from a symptom list. Counting how often ignition complaints say “shaking” establishes nothing, because owners say shaking about everything. What matters is whether they say it more here than elsewhere.

So every symptom below is counted twice: once across the 1,082 ignition-naming engine complaints, and once across the other 9,946 engine complaints in the same pull. Both groups come from the same vehicles, the same endpoint and the same retrieval, which means they carry the same biases — the same reporting culture, the same model mix, the same self-selection, the same tendency for a fault that made the news to be over-reported. Those biases cancel in the comparison. The absolute rate in either group is close to meaningless. The gap between them is not.

Ten symptom categories were defined in advance, each as a list of the actual words and phrases people use, and each is printed in the data file beside its counts so anybody can see exactly what was counted as what. “Rough running” catches rough idle, runs rough, sputtering, stuttering and chugging. “Shaking” catches vibration, shudder, judder, jerking and bucking. Nothing was tuned after the fact to make a result come out.

The three symptoms that point at ignition

Read the two columns as a pair and the list sorts itself into three groups: symptoms that run well above the background rate, symptoms that run at about the background rate, and symptoms that run below it.

Rough running is the strongest single signal in the file. It appears in 128 of the ignition narratives, 11.8 per cent, against 3.2 per cent of the other engine complaints. Nothing else on the list separates the two groups by that margin. If an owner writes that the engine runs rough, sputters, stutters or chugs, the odds that the narrative also involves ignition are close to four times what they would be otherwise.

That fits the mechanism, which is the reassuring part. A cylinder that fails to burn its charge does not degrade the engine’s output evenly; it removes one contribution from a rotating sequence and leaves the others intact. The result is not a weaker engine, it is an uneven one, and unevenness is exactly what “rough” describes. An engine losing power for almost any other reason — a restricted intake, a tired pump, a clogged exhaust — tends to lose it smoothly.

Shaking, vibration and jerking run at 17 per cent against 8.3 per cent, roughly double. That is the same physical fact reaching the driver through a different sense. A cylinder that misses puts a gap in the torque pulse going through the crankshaft, the flywheel and the mounts, and that gap is felt in the steering wheel and the seat before it is felt as a change in speed. Our page on a car that shakes while driving works through the other candidates, most of which are wheels, tyres and driveline rather than ignition — but a shake that appears with a warning light and gets worse under load belongs in this file rather than that one.

Loss of power or limp mode runs at 19.6 per cent against 11.8. It is a real signal but a much weaker one than the first two, and that makes sense: reduced power is the final common path for a very long list of faults, and a modern engine management system will deliberately cut output for reasons that have nothing to do with the ignition system. Hearing it should raise your suspicion a little and settle nothing.

Two smaller ones are worth a line. Smoke or a smell of fuel appears at 10.4 per cent against 7.5, a modest lift; a cylinder that does not light sends its fuel out through the exhaust valve, so an unburnt-fuel smell is mechanically consistent, but it is not distinctive enough to lean on. Hesitation and stumbling appears in only 46 narratives, 4.3 per cent against 3.3, which is a lift too small to argue about at that sample size. Worse fuel economy appears in 11 narratives out of 1,082 — 1 per cent against 0.9 — and is, on this evidence, simply not something people report. It is real and it is invisible, which is the worst combination a symptom can have.

The three symptoms that point away

This is the part of the measurement that a symptom list can never produce, because a list has no way to say that an item on it is evidence against the thing it is listed under.

Hard starting or no start is the clearest reversal. It appears in 3 per cent of ignition narratives against 6.2 per cent of the other engine complaints — half the rate. It is near the top of almost every published list of bad-plug symptoms, and in this file it is a reason to look elsewhere first.

The mechanism explains why. Starting a cold engine is not the hardest thing you can ask a spark plug to do; it is one of the easier ones. Cylinder pressure at cranking speed is low, and the voltage a spark needs to jump a gap rises with the pressure it is jumping through. A worn plug with a widened gap runs out of margin under load first — pulling hard, climbing, accelerating — and keeps working at idle and at cranking long after. An engine that will not start has usually lost something more fundamental than spark quality on one cylinder, and the file behaves accordingly.

Knocking, ticking and rattling runs at 6.6 per cent against 9.5. Mechanical noise is a bearing, a lifter, a chain, a heat shield or a valve. It is not what a marginal spark looks like from the driving seat, and its under-representation here is a small mercy: a knock in an engine is usually a more serious finding than anything on this page, and it should not be filed under ignition simply because a list said it could be.

Stalling runs at 15 per cent against 17.7, slightly below the background rate. This one deserves care rather than dismissal. Stalling absolutely can come from ignition trouble, and 162 narratives in this group describe it. What the comparison says is only that it does not distinguish ignition trouble — an engine can stop for so many reasons that hearing “it stalled” moves you no closer to the cause. It is a symptom that demands attention and gives no direction.

And the dashboard light sits in a category of its own. At 57.7 per cent it is by far the most-named item in the ignition group, appearing in 624 narratives, but it appears in 28.5 per cent of the other engine complaints too. It is doing exactly what it was designed to do: telling you that the engine management system is unhappy, without telling you why. Our page on what a check engine light actually means covers what to do about that, including the distinction between a steady light and a flashing one, which is the single most consequential thing on this subject.

Which came first, the feeling or the light?

Complaint narratives are broadly chronological. People describe what happened in the order it happened, which means the order symptoms appear in the text is usable evidence about the order they were noticed. That is a slightly unusual thing to be able to measure, and it answers a question the symptom lists never even ask: when this failure announces itself, what does it announce itself as?

Not every narrative can answer. 218 of the 1,082 — 20.1 per cent — name no driver-side symptom at all; they describe the failure entirely in workshop language, a diagnosis without a story. Another 385, 35.6 per cent, name exactly one thing. That leaves 479 narratives, 44.3 per cent, that name two or more symptoms and therefore contain an order. The mean across the whole group is 1.46 symptom categories per narrative.

Ranking those 479 by which symptom opens the account gives a first answer: shaking opens it 49.1 per cent of the times it appears alongside something else, rough running 47 per cent, the warning light 44.5 per cent. But that comparison mixes narratives containing different combinations of symptoms, so a cleaner version is a straight two-way contest — among narratives that name both a given feeling and a warning light, which one is written down first.

The results split the symptoms into two kinds, and the split is genuinely useful.

Some feelings arrive before the light. Of the 124 narratives naming both shaking and a warning light, the shaking is named first in 73 of them, 58.9 per cent. Rough running beats the light in 45 of 85, 52.9 per cent. Hesitation beats it in 22 of 36, 61.1 per cent, on a small sample. These are the symptoms a driver can register while the engine management system is still deciding whether anything is wrong — a misfire has to persist and be counted before a code sets, and a person feels the roughness during the counting.

Some events arrive after it. Of the 99 narratives naming both stalling and a warning light, the light is named first in 58.6 per cent. Of the 69 naming both a fuel smell or smoke and a light, the light comes first 71 per cent of the time. Loss of power is close to even at 48.1 per cent. These are not early warnings; they are what happens next, to a car whose light has been on for a while.

One confound has to be handled here rather than hidden. About a fifth of these narratives — 224 of 1,082, or 20.7 per cent — open with the call handler’s stock sentence rather than the owner’s own words, and a form has an order of its own. So the same contest was run on the 858 owner-written narratives alone. Shaking still beats the light, 50.6 per cent of 87. Rough running still beats it, 51.6 per cent of 62. Stalling still loses to it, 39.2 per cent of 79. Smoke still loses heavily, 23.4 per cent of 47. The direction holds on every result the page leans on, which is the test that matters.

The practical reading is short. Waiting for the light is waiting too long, and the light coming on after you already felt something is the ordinary sequence rather than a sign that the fault suddenly worsened. If you first noticed a roughness and the light appeared a week later, those are the same event.

What each symptom is also produced by

None of the above changes the fundamental problem, which is that a symptom is a shared output. The table below is the one a symptom list ought to print and never does: for each item, the other things that produce it, and the observation that actually separates them. Nothing here requires taking anything apart.

The row that does the most work is the first one. A misfire that a scan tool can pin to a specific cylinder is a much narrower problem than a misfire it reports as random, and the difference is diagnostic rather than cosmetic. Our page on engine misfires works through that split properly, including what a random-cylinder code usually means and why it is a different investigation.

The mileage this arrives at, and why it matters before anything else

There is a prior question that should come before any symptom, and the complaint file happens to answer it. Roughly a fifth of these records — 236 of the 1,082, 21.8 per cent — carry an odometer reading in the intake template, written by the person taking the report rather than by the owner in prose. That is the only mileage figure in these records that can be parsed without inventing an interpretation, so it is the only one used here.

The median across those 236 readings is 93,635 miles. The lower quartile is 73,600 and the upper quartile is 114,000. Only 13.6 per cent of them sit below 60,000 miles, and 59.3 per cent below 100,000.

Set that against the same calculation on the other engine complaints, and the interesting part is not the middle but the bottom. The 2,163 odometer readings on other engine complaints have a median of 87,527, which is not far off — but their lower quartile is 58,000 against 73,600, and 25.8 per cent of them fall below 60,000 miles against 13.6 per cent. Ordinary engine trouble in this file happens across a broad span of mileages including a great many young cars. Ignition trouble is concentrated further up the odometer and starts later. Across every complaint in the pull, whatever the component, the 7,434 readings have a median of 74,000.

That gives you a prior, and a prior is worth more than a symptom. If a car is showing every item in the table above at 30,000 miles, the plugs are very unlikely to be the answer — nothing about them has changed yet, and something else is producing the same output. If the same car is showing them at 110,000 with no record of the work ever having been done, the plugs move to the front of the queue on mileage alone, before a single symptom is weighed. Our guide to what a plug change involves and how to find out whether it has been done is the other half of that question: the interval, what decides the size of the job on a particular engine, and how to search a service folder for work that is nearly always bundled inside something larger.

One limitation, stated plainly because it is severe. This is right-censored data. These are the 2012 to 2018 model years observed in 2026, so relatively few of these cars have yet been driven far past 200,000 miles, and the thin upper tail is a fact about the fleet’s age rather than evidence that failures stop. Readings below the median are sound — censoring can add events later, it cannot remove them from below — which is why the lower quartile and the share below 60,000 are the figures this page leans on and the upper tail is not.

When an owner names a spark plug, what else is in the sentence

Now the finding that changed how this page is written.

557 narratives in the pull name a spark plug. Reading what else those same narratives name is sobering. 226 of them, 40.6 per cent, also name a misfire, which is expected. But 152, 27.3 per cent, name a recall or a technical service bulletin. 91, 16.3 per cent, name a replacement or rebuilt engine. 90, 16.2 per cent, name a head gasket. 88, 15.8 per cent, name oil consumption or burning oil. 59, 10.6 per cent, name fuel injectors. 28 name a valve cover, valve seal, piston ring or the crankcase ventilation system. 24 describe a fouled plug. Only 8 name the catalytic converter and 6 a timing chain or belt.

Put that together and a pattern emerges that no symptom list would ever produce. In the federal complaint file, a spark plug is very often not the subject of the complaint. It is the thing that was found, or replaced, or fouled, on the way to something considerably worse. An engine consuming oil fouls its plugs; a head gasket letting go contaminates a cylinder; an injector delivering badly leaves a plug looking like the culprit. In each of those the plug is the messenger.

This is a selection effect and it should be named as one. Nobody files a federal safety complaint because their plugs reached the end of their interval, so the plugs that reach this archive are disproportionately the ones that were caught up in something serious. The file is a biased window by construction. But the bias is itself the lesson for anybody standing next to a car with these symptoms: the population of “spark plug problems” that gets written about publicly is not the population of ordinary plug wear, and a plug found fouled is a finding that needs a cause attached to it rather than a job that is finished when the new one goes in.

The practical consequence is a question. If a shop tells you the plugs were fouled, the useful follow-up is not what they cost. It is why — whether all of them looked the same or one looked different from the rest, and what the deposit was. A plug that comes out wet with oil is reporting on where oil is getting in, and replacing it treats the report rather than the thing being reported.

What the file does not show, which is as useful as what it does

Three measurements came back empty or close to it, and they are worth printing because an absence stated is more honest than an absence quietly dropped.

Circumstance did not separate the two groups at all. The received wisdom is that a marginal plug fails under load — accelerating, climbing, overtaking — and holds up at idle. The mechanism is sound and this page has already leaned on it. The complaint text does not corroborate it. Highway or freeway speed appears in 24.4 per cent of ignition narratives against 25.7 per cent of other engine complaints. Accelerating, merging or pulling away appears in 14.7 per cent against 16. Idling or stopped in traffic appears in 7.1 per cent of both. Cold weather, load and rain are each named in under 3 per cent of either group. The honest conclusion is that people describing a fault to a government agency write about circumstance in a fairly standard way regardless of the fault, and this dataset is the wrong instrument for that question. It is not evidence against the load explanation. It is no evidence either way.

The severity flags run the other direction from everything else on the file. Ignition complaints carry a crash flag 0.2 per cent of the time against 1.6 per cent for other engine complaints, and a fire flag 0.3 per cent against 4.8 per cent. They lead to a tow 13.8 per cent of the time against 16.6, and to the driver pulling over or being stranded 10 per cent against 11.5. This is a driveability failure, not a safety event, which is precisely why so little of it reaches a safety complaint database in the first place. It is also why the absolute counts on this page mean so much less than the ratios.

What ignition complaints do have is answers. 40.4 per cent of them name a diagnosis against 23.7 per cent of the other engine complaints — by a wide margin the largest gap in the outcome measures. Whatever else is true of this failure, it is one that shops identify. Set against that, 15.7 per cent of ignition complaints say the vehicle was not repaired, against 11.9 per cent elsewhere; a diagnosis is not the same as a resolution, and a meaningful number of these people knew exactly what was wrong and did not get it fixed.

And one thing the file cannot contain at all, which bears repeating because it governs everything above: there is no maintenance information in it. No service dates, no part numbers, no record of what was fitted or when. A spark plug wearing out on schedule and being replaced on schedule is invisible here and always will be. Everything on this page is a measurement of complaints, not of plugs.

What to do with a car showing these symptoms

Common questions

What are the most reliable symptoms of a bad spark plug?

Measured against a control group of other engine complaints from the same federal pull, rough running is the strongest: 11.8 per cent of ignition narratives against 3.2 per cent of the rest. Shaking, vibration or jerking is second at 17 per cent against 8.3. Loss of power is third but much weaker at 19.6 against 11.8. A dashboard warning light is the most commonly named item of all at 57.7 per cent, but it appears in 28.5 per cent of other engine complaints too, so it confirms that something is wrong without narrowing what.

Can a bad spark plug stop a car from starting?

It can, but it is not the way to bet, and the complaint file is unusually clear on this. Hard starting or no start appears in 3 per cent of ignition-naming engine complaints against 6.2 per cent of other engine complaints — half the rate. Starting is one of the easier jobs you can give a plug, because cylinder pressure at cranking speed is low and the voltage a spark needs rises with the pressure it jumps through. A worn plug therefore fails under load long before it fails at cranking. If an engine cranks normally and will not fire, look at fuel, at the crank position sensor and at the immobiliser before the plugs.

Will the check engine light always come on?

Not necessarily first, and that is the more useful answer. In narratives that name both a warning light and shaking, the shaking is written down first 58.9 per cent of the time; for rough running it is 52.9 per cent. A misfire has to persist and be counted before the engine management system sets a code, and a driver can feel the roughness during the counting. Waiting for the light is waiting past the first evidence. Conversely, stalling and a fuel smell are usually named after the light — 58.6 per cent and 71 per cent respectively — which means they are what happens next rather than early warnings.

At what mileage do spark plug problems usually show up?

In this pull, the median odometer reading on an ignition complaint that records one is 93,635 miles, with a lower quartile of 73,600 and an upper quartile of 114,000. Only 13.6 per cent are below 60,000 miles, against 25.8 per cent of other engine complaints. That is a real difference in shape rather than a small one: ordinary engine trouble is spread across the mileage range including plenty of young cars, while ignition trouble starts later and clusters higher. Use it as a prior. On a low-mileage car the same symptoms usually point somewhere else.

How do I tell a bad spark plug from a bad ignition coil?

You mostly cannot, from the driving seat, and the complaint file shows that people rarely try — only 142 narratives across the whole pull name a coil or coil pack at all, against 557 naming a plug. The two produce the same symptom because they are two halves of the same job, and the distinction is made with a scan tool rather than a description: a fault that follows a coil when it is moved to another cylinder is the coil, and a fault that stays with the cylinder is the cylinder or the plug in it. This is one of the reasons the two parts are so often replaced together.

Is a spark plug problem dangerous?

Rarely, in the sense that safety databases measure. Ignition-naming engine complaints carry a crash flag 0.2 per cent of the time against 1.6 per cent for other engine complaints, and a fire flag 0.3 per cent against 4.8 per cent. They are also slightly less likely to end in a tow, 13.8 per cent against 16.6. The exception is the one that matters: a heavy, sustained misfire pushes unburnt fuel into the catalytic converter, where it burns at temperatures the converter was never built for. That is what a flashing warning light is telling you, and it is the one situation on this page where continuing to drive converts a small problem into an expensive one.

If my plugs were fouled, is replacing them the end of it?

Not on this evidence. Of the 557 narratives naming a spark plug, 16.2 per cent also name a head gasket, 15.8 per cent name oil consumption or burning oil, 16.3 per cent name a replacement or rebuilt engine and 10.6 per cent name fuel injectors. A fouled plug is a record of what has been arriving in that cylinder, so the question is what fouled it and whether every plug looked the same. This is a biased sample — nobody complains to the federal government about plugs wearing out normally — but the bias runs in the direction of caution, which is the right direction when a shop hands you a plug that is wet with oil.

Does this data prove worn spark plugs cause these symptoms?

No, and it is important to say so. Membership of the group measured here means only that a narrative names a spark plug, an ignition coil or a misfire — not that a plug was faulty, and not that a plug caused anything. Complaints are self-selected, unverified, and skewed towards vehicles and faults that made the news. The file also holds no maintenance information at all, so a plug reaching the end of its life on schedule never appears in it. What the comparison supports is a statement about language: which symptoms owners write down when ignition hardware is in the story, relative to when it is not. That is a genuinely useful thing to know, and it is less than a causal claim.