How many lithium-ion battery fires are there?

The report is NIST Technical Note 2365, "Understanding the Risk of Lithium-Ion Battery Fires: multi-source data analysis," by Stanley Gilbert and five colleagues from NIST's Applied Economics Office and Fire Research Division, published 10 March 2026. It is free to read, and it is unusually frank. Its abstract says data on these fires "is fragmented and mostly incomplete," and its conclusion is that current reporting substantially underestimates the problem.

If you own a home battery, or are thinking about one, this is worth understanding. Every confident statement you hear about how often batteries catch fire, in either direction, is built on data this report takes apart.

Why the fire records miss most of them

The main US source is NFIRS, the National Fire Incident Reporting System, where fire departments file reports on the incidents they respond to. The report lists three problems with using it to count battery fires.

Most fires never reach it. Participation is voluntary. NIST cites the NFPA estimate that of 1.39 million fires reported to fire departments in 2023, about 644,000 made it into NFIRS. Many fires are never reported to a fire department at all.

It was not built for batteries. NFIRS predates common lithium-ion use and has no field that identifies a lithium-ion fire. The US Fire Administration recommended a consistent way to code them, and NIST's analysis found it is not used consistently.

Departments code them very differently. NIST looked at 23 large departments from 2019 to 2023. The number of structure fires per identified battery fire ranged from 120 at West Metro Fire and Rescue in the Denver area to more than 6,000 at the Chicago Fire Department. The report reads that spread as differences in identification, not in how many battery fires actually happen.

Out of about 2.66 million structure fires reported to NFIRS from 2011 to 2023, 2,063 were identified as battery fires, roughly 0.08 percent.

How you count what nobody counted

NIST's answer is a method borrowed from wildlife biology, called capture-recapture. Catch and tag some fish, release them, catch again later, and see how many of the second catch already carry tags. If only a few do, the lake must hold far more fish than you have seen.

Here the two "catches" are NFIRS and the Consumer Product Safety Commission's SaferProducts database, where the public files complaints about products. NIST pulled about 3,800 CPSC records mentioning batteries or lithium and used a large language model (Llama 4 Maverick, each question asked ten times per record, with disagreements thrown out) to pick out lithium-ion incidents that involved a fire. That left 557.

Then the team looked for fires that appear in both lists: same city and state, with the consumer complaint filed no more than 7 days before or 90 days after the fire report. That turned up 16 candidates. Reviewed by hand, 5 held up.

Five. That is the number the whole estimate rests on. If only 5 of the 557 consumer-reported fires show up in NFIRS, the fire records are catching something like 1 in 100 of these fires (that ratio is my arithmetic from the published counts). NIST's statistical model, fitted with the Stan software, puts the total at about 198,000 lithium-ion fires in US structures from 2011 to 2023, with a 95 percent interval of 84,000 to 465,000.

Why that range is so wide

The textbook formula is simple: multiply the two list sizes and divide by the overlap. With these counts, 2,063 times 557 divided by 5 gives about 230,000, in the same neighborhood as NIST's model. Now change the overlap by one. Six matches would give about 192,000. Four would give about 287,000. (Those sensitivity numbers are mine, using the simple formula, not NIST's model.) When the denominator is 5, one fire more or less moves the answer by tens of thousands. That is why the interval spans more than a factor of five, and NIST reports it honestly.

There is a second, quieter assumption. The method treats the two lists as independent, and NIST argues they are, since they are collected in very different ways. My read: if bigger, more damaging fires are more likely to land in both lists, the overlap gets inflated and the true count would be even higher than the estimate. NIST does flag a related bias elsewhere in the report, noting that identification of battery fires is probably not independent of whether someone was hurt.

What the report says about home fires

With the caveat that the fires NFIRS did catch may not represent the ones it missed, NIST found that battery fires in homes peak around 6 p.m., are most common in summer, and most often start in the garage, followed by bedrooms and living areas. The report notes the garage figure may include lead-acid batteries. More than 80 percent of home battery fires spread beyond the object where they started. Consumer lithium-ion fires in the CPSC data are growing about 10 percent a year (confidence interval 7.1 to 13.0).

For vehicles, NIST paired NFIRS with a crowd-compiled Tesla fire list and estimated about 5,718 electric and plug-in hybrid car fires since 2011, excluding Florida, whose unusually high match rate the authors think skews results. Only about 1 in 10 of those fires is identified in NFIRS by vehicle identification number, against about 1 in 3 for car fires overall.

What it does and does not say about home batteries

This is the part people will get wrong. The 198,000 figure is every kind of lithium-ion fire in a structure: phones, laptops, vapes, e-bikes, scooters, power tools. NIST's review of news coverage points to e-bikes and scooters as a leading cause of home battery fires. It is not a home energy storage number, and the report does not produce one.

On stationary storage, the report has two small data points. In UL's incident database for 2024, energy storage systems were less than 1 percent of incidents (33). EPRI's battery storage failure database averaged 2.93 incidents per quarter from 2018 to 2024 with no statistically significant trend. NIST is careful about both: UL's database is a sample built mostly from news reports, and EPRI's completeness is listed as unknown, because it also depends on what gets reported publicly. EPRI's database is also the source behind the falling failure rate in Report 083, which is a rate per deployed gigawatt-hour and not a count.

So the honest takeaway for a homeowner is not "batteries are burning everywhere," and not "home batteries never burn." It is that the official counts are a floor, not a measurement. The practical defenses do not depend on the count: a listed system installed to code (Report 101 covers what fire testing does and does not certify), smoke alarms where it lives, and keeping e-bike and scooter charging out of bedrooms and exits.

What fixes this

NFIRS is being replaced by NERIS, the National Emergency Response Information System, which is rolling out through 2025 and 2026 and is designed to identify lithium-ion fires and record details about them. NIST says there is not yet enough NERIS data to analyze. Once there is, a count that today rests on five matched fires could rest on actual coded reports. If a NERIS-based count comes out that contradicts these numbers, I will update this report in place.

A note on my angle: I help design the AI battery-cycling systems for a veteran-owned (HUBZone) energy-storage integrator, run by fellow veterans. I do not own that company and earn nothing from this link. Full policy here. Nothing in this report involves them, and the NIST data does not single out any product or brand of home battery.