Every large medical site answers this with an adjective. Mayo says safe and well tolerated, Yale says as safe as an MRI, Cleveland Clinic says safe. None of them is wrong. What none of them gives you is the number, who that number belongs to, or what the device label actually says. This page does that, and says where the evidence runs out.
The short answer
Serious harm from TMS is rare and the rarity is well documented. In the largest published count, 18 seizures were reported across 586,656 treatment sessions. But the risk is not spread evenly across the people sitting in the chair. In a separate survey of 318,560 sessions, sessions delivered inside the published guidelines to people with no identified risk factors produced 4 seizures. Sessions in people who did have risk factors produced 19 — roughly sixteen times the rate, from the same machines. Almost all of the risk is decided by screening, coil choice and dose, before the first pulse is delivered.
What the top results say
Source
What it says about safety
Mayo Clinic
"rTMS is considered safe and well-tolerated"
NIMH
"the safety profile of TMS is excellent"
Yale Medicine
"TMS is as safe as having an MRI"
Cleveland Clinic
"It's safe" and "very rarely causes seizures"
Google's AI summary
"safe, non-invasive, and FDA-approved"
Three of those five statements carry no number at all, and the fifth contains a factual error. TMS devices are not FDA approved. They are cleared, and the first one was not even that. The same summary puts seizure risk at "less than 0.01% of sessions" without naming a study, which matters, because the published figures do not agree with each other and the way they disagree is the most useful thing on this page.
This page is not here to argue that TMS is dangerous. The evidence says it is not, for most people, most of the time. It is here because "safe" is a conclusion, and you are entitled to the working.
The serious risk
24 in 318,560
Seizures reported across five years of sessions at 174 labs and clinics
18 in 586,656
Seizures reported in 25,526 patients, Clinical TMS Society survey
62%
Share of seizures that happened on someone's first exposure to TMS
Those work out to 0.08 seizures per 1,000 sessions in the first study and 0.31 per 10,000 sessions in the second. The second also reports it per person, which is closer to the question you are actually asking: 0.71 seizures per 1,000 patients treated.
Both are voluntary surveys rather than registries, and both say so. The first drew 174 responses from a possible population in the thousands, and its authors call their own data "only semi-quantitative"; 63% of the sites estimated their session counts rather than counting them. The second surveyed the membership of the Clinical TMS Society and got roughly a 27% response, and could not independently verify the seizures reported to it. They agree with each other to within about a factor of three, which for two self-report surveys is reassuring, and is still not the same thing as a registry.
One number is worth tracing, because you will see it everywhere. Clinic pages routinely give the risk as fewer than 1 in 30,000 treatments and attribute it to the international safety guidelines. It does not appear in those guidelines. It comes from the NeuroStar device labeling, which states it without a denominator or a published derivation. The guidelines’ own summary, in their 2021 update, is that the risk "is about <0.03%" — three in ten thousand, an order of magnitude higher than the label figure, and also given without a denominator. Three numbers, one question, and the gap between them is not resolved anywhere.
Fainting is far more common than a seizure and looks like one from the outside. About 17% of the sites surveyed in 2019 had seen at least one patient faint, and the 2021 guidelines specifically ask clinicians to tell a seizure apart from convulsive syncope. One manufacturer's post-market monitoring counts the two together, which is worth knowing before comparing its figures with anyone else's.
The finding
The overall rate is the least useful way to read this data. The same survey, split by who was in the chair and what protocol was used, says something you can act on.
All sessions
0.08 · 24 seizures in 318,560 sessions
No elevated risk of either kind
0.02 · 4 seizures in 242,067 sessions
Elevated protocol risk only
0.00 · 0 seizures in 12,559 sessions
Elevated patient risk
0.33 · 19 seizures in 57,185 sessions
Elevated patient and protocol risk
0.15 · 1 seizure in 6,749 sessions
Nineteen of the twenty-four seizures happened in the 18% of sessions delivered to people who already had a recognized risk factor. Four happened in the 76% of sessions delivered to people who did not. Not one happened in the 12,559 sessions where an unusual protocol was used on somebody without risk factors.
The first is you. A history of seizures, stroke, head injury, a neurological condition that lowers the seizure threshold, or a medication that does the same. The second is the coil. In the 2019 survey, conventional figure-of-eight and round coils ran at 0.08 per 1,000 sessions, double-cone coils at 0.12 and H-coils at 0.43. In the 2021 survey, the H-coil device came in at 5.56 seizures per 1,000 patients against 0.14 for the three most widely used figure-of-eight devices. The guideline authors are careful here: they do not have comparable data from every manufacturer, so this is a signal rather than a settled ranking. The third is when in the course you are. Over 62% of seizures happened on a first exposure and 75% within the first three. If you have had TMS before without incident, you are in a lower-risk group than you were on day one.
Which is why the honest answer to "is TMS safe" is a question back. Safe for whom, with which coil, at what dose, screened by whom. The rest of this page is about how to get those four answers.
Regulatory status
Approval is the pathway for high-risk devices, and it requires a manufacturer to show on its own evidence that the device is safe and effective. Clearance is the pathway for moderate-risk devices, and it requires the manufacturer to show the device is substantially equivalent to something already on the market. Almost every TMS system in a US clinic today is cleared, not approved.
Device
How it was authorized
Date
NeuroStar (Neuronetics)
De Novo classification granted, DEN070003
7 Oct 2008
BrainsWay Deep TMS
510(k) cleared, K122288
7 Jan 2013
BrainsWay Deep TMS for OCD
De Novo, marketing permitted
17 Aug 2018
Magnus with SAINT
510(k) cleared, K220177
1 Sep 2022
The first one is not a clearance at all. Neuronetics filed a 510(k), the FDA found the device not substantially equivalent to anything already being sold, and it was then classified into Class II through the de novo route in October 2008. That order created the regulatory category, 21 CFR 882.5805, that every TMS device since has used as its predicate. The entire US TMS market descends from one device that could not clear the ordinary route.
The category's own definition is narrower than the way TMS gets talked about. It describes a device for major depressive disorder in patients "who have failed at least one antidepressant medication and are currently not on any antidepressant therapy." Everything else — OCD, anxiety, smoking cessation, and every off-label use — sits outside that original clearance and either has its own authorization or has none.
Two sentences from the NeuroStar labeling are worth reading before anyone tells you TMS is proven. "The acute effectiveness of NeuroStar TMS Therapy has not been established beyond a six-week treatment course for MDD." And: "This device has not been evaluated for durability of antidepressant effect in controlled clinical trials."
TMS systems are prescription devices by regulation, and the labeling requires an operator trained to recognize and manage a seizure to be in the room for the whole session. There are 66 device records under the FDA's transcranial magnetic stimulator product code and not one of them is a home or consumer device. Nothing on this page applies to anything sold for home use under the name TMS.
Who should not have it
From the NeuroStar device labeling. These two are the only contraindications it lists.
Conductive, ferromagnetic or magnetic-sensitive metal implanted within 30 cm of the coil: aneurysm clips and coils, stents, implanted electrodes, bullet fragments. Removable things — jewelry, hair clips — simply come off before a session.
Including their leads. Deep brain stimulators, cochlear implants, vagus nerve stimulators. The 2021 guidelines are blunt about one of them: people with cochlear implants should not undergo TMS.
Standard amalgam fillings, single-post dental implants and bridge work.
Any implant that is activated or controlled by physiologic signals, including pacemakers and implantable or wearable defibrillators, even outside the 30 cm distance.
The label's special-populations list, which is the exclusion criteria of the studies that got the device cleared.
The right-hand column is not a list of people TMS is known to harm. It is a list of people it was never tested on. Some of them are treated with TMS every day, off-label and reasonably. What the label is telling you is narrower and more useful than a warning: if you are in that column, the numbers on the rest of this page were not measured on anyone like you.
The international guidelines put it more sharply. The only absolute contraindication they recognize is metal hardware in close contact with the coil. A seizure history or metal inside the skull is, in their words, "typically exclusionary but not absolutely contraindicated" — a risk-benefit judgment for the treating physician rather than a locked door.
What actually happens
| Reported effect | Active TMS, n=165 | Sham, n=158 |
|---|---|---|
| Headache | 58.2% | 55.1% |
| Pain at the treatment site | 35.8% | 3.8% |
| Muscle twitching | 20.6% | 3.2% |
| Discomfort at the treatment site | 10.9% | 1.3% |
| Skin pain | 8.5% | 0.6% |
| Toothache | 7.3% | 0.6% |
| Facial pain | 6.7% | 3.2% |
| Eye pain | 6.1% | 1.9% |
Headache is the effect everybody warns you about and it is the one the sham column clears. Fifty-eight percent of people on real TMS got a headache and fifty-five percent of people under a fake coil got one too. Pain where the coil sits is the real one: nine times the sham rate, and the reason the labeling admits that "unblinding due to treatment related adverse events, such as application site pain, was a concern in this study" and that the sensitivity analyses for it "were not conclusive."
Almost nobody stopped. All-cause discontinuation was 7.7% on active TMS against 8.2% on sham, and discontinuation because of a side effect was 4.5% against 3.4% — a lower dropout rate than the sham group in one case and a near-identical one in the other. There were no deaths and no seizures in the controlled trials, and none in the more than 10,000 sessions run before clearance.
Two device-related serious events did happen, and they are worth knowing because they tend to get rounded away. Both were device malfunctions: two first-degree burns and one episode of severe pain at the treatment site, all classed as probably related to the device. Current systems warn the operator when the coil surface may exceed 41°C. If your scalp is burning, that is not something to sit through politely.
Treatment-emergent mania is the other effect people ask about. Across 53 randomized trials in depression there were 13 cases — 0.84% on active rTMS against 0.73% on sham, and below the 2.3 to 3.45% switch rate seen in people with bipolar disorder taking mood stabilizers alone.
Read this part first
The FDA requires every TMS device label to carry a warning, in bold, about depression worsening and about suicidal thinking emerging during treatment. That requirement stands whether or not the device causes it — and in the registration trial the evidence pointed the other way.
Six people were hospitalized for worsening depression during the study. None of them were receiving active TMS: four were in the run-in phase before randomization and two were on sham. Seven were hospitalized after suicidal thinking or an attempt — two in the run-in phase, two on active TMS and three on sham. The one suicide attempt was in the sham group. Every one of these events was adjudicated as not related to the device.
Severe depression carries this risk on its own, and a course of TMS is six weeks of it. The warning is about the illness rather than the coil, which is why the same bolded language sits on antidepressant labeling. What it means in practice is that somebody should be watching. Being seen five days a week is one of the few things about TMS that makes this easier rather than harder — the technician you see every morning is a person you can tell.
If you are thinking about harming yourself, use one of these now.
988 Suicide & Crisis Lifeline
Call or text 988, or chat at chat.988lifeline.org. Free, confidential, 24 hours a day. Press 2 for Spanish, or text the word Ayuda. Press 1 if you are a veteran or service member. Deaf and hard-of-hearing callers can reach 988 by videophone, or dial 711 then 988 through a relay service.
Crisis Text Line
Text HOME to 741741. In Spanish, text HOLA or AYUDA to the same number.
The one nobody mentions
The coil clicks because it physically deforms when it discharges. Measured carefully, a single pulse peaks at around 136 to 139 decibels, and continuous stimulation at full output has been measured between 96.5 and 112 decibels A-weighted. Those figures sit above occupational noise exposure limits. They are also hard to measure — the pulse is short enough that a standard sound meter can underestimate the peak by as much as 50 decibels, and the coil is resting against your skull, so some of the sound arrives by bone conduction and is not captured at all.
With hearing protection this appears to be a solved problem. The registration trial found no change in hearing thresholds at four or six weeks with 30 dB protection in place, and the device labeling requires earplugs for the patient and the operator at every single session. The one documented case of permanent hearing change in this literature is a person whose earplug slipped out of one ear partway through stimulation.
Which makes this the rare risk where the whole mitigation is in the room with you. The guidelines note that foam earplugs are difficult for many people to fit properly on their own. Ask someone to check yours, and say something the moment one works loose.
Memory and cognition
In the registration trial, cognition was tested at baseline, four weeks and six weeks using the Mini Mental State Examination, the Buschke Selective Reminding Test and the Autobiographical Memory Interview. No change was found on any of them. A later systematic review of 31 randomized controlled trials of prefrontal rTMS, across depression, schizophrenia and Alzheimer's disease, found no reliable evidence of cognitive side effects — with a handful of studies showing small improvements in attention and executive function instead.
For contrast, and because it is the honest version of the comparison rather than the scary one: a meta-analysis of 84 studies and 2,981 people who had ECT found a measurable drop across most cognitive measures in the first three days afterwards. By four to fifteen days, every measure but one had returned to baseline or above. Beyond fifteen days there was no remaining deficit at all, and 57% of measures had improved past where they started. ECT does have a cognitive cost, it is real, and it is mostly a matter of days. TMS does not appear to have one.
The caveat the guideline authors make about their own field is worth repeating. Cognitive outcomes in rTMS trials are usually exploratory analyses buried inside large test batteries, rarely pre-registered, and inconsistently corrected for false positives. "No reliable evidence of harm" is a fair reading of this literature. "Proven to have no cognitive effect" is not.
The limits
The longest systematic follow-up is a naturalistic study of 257 people over 52 weeks, which found no medically serious device-related events and no suicides. Everyone in it had already completed a course and volunteered to continue, so it describes people for whom TMS went well. There is no controlled safety study beyond twelve months.
Two voluntary surveys, response rates of roughly 27% and unknown, session counts estimated rather than counted at most sites, and no independent verification of the events reported. The authors of the first call their own data only semi-quantitative.
Under 22, over 70, bipolar disorder, psychotic features, a seizure history, a recent suicide attempt, active substance use and pregnancy were all excluded from the studies these numbers come from.
Pain at the treatment site is nine times more common on real TMS, which tells a fair number of people which group they are in. The labeling says the sensitivity analyses for this were not conclusive.
The guidelines define low risk as stimulation inside published dose limits with a conventional figure-of-eight coil. Accelerated schedules, theta-burst and deep coils are newer and less well counted, and are not automatically covered by the figure-of-eight safety record.
The FDA adverse event database holds 338 reports for these devices, 71 of which mention a seizure. The FDA says plainly that the data is not intended to evaluate rates or to compare devices: there is under-reporting, no verified causality, and no denominator.
Before you start
The evidence on this page says most of the risk is set by decisions taken before you sit down. These are the questions that reach those decisions. None of them is confrontational and every one of them has a right answer.
Seizures, epilepsy, stroke, head injury, and any medication that lowers the seizure threshold. Ask which of your medications they checked against that list.
Figure-of-eight, double-cone or H-coil. The published seizure rates differ between them, and you are entitled to know which one is going over your head.
And how often the motor threshold is re-measured. It moves over a course, and a stale threshold is a dosing error in either direction.
Standard daily stimulation for depression sits inside the original clearance. Accelerated schedules, theta-burst, and treatment for anything other than depression may not.
The labeling requires an operator trained to recognize and manage a seizure, present for the whole session rather than just settling you in.
A clinic that answers this quickly and specifically has thought about it. One that looks surprised by the question has not.
And who checks them, given the guidelines note that most people cannot fit foam plugs properly on their own.
Not the front desk. The clinician who can actually change something.
FAQ
Yes, and they divide into three kinds. Discomfort, which is real: pain where the coil sits was reported by about a third of people in the trial and usually settles within the first week or two. Time, which is the underrated one: roughly 36 sessions, five days a week, most of which you have to travel to. And the possibility that it does not work, which is far more likely than any physical harm described on this page. The serious physical risks are genuinely rare.
There is no evidence that it does. The pulse induces electrical activity in the outer few centimeters of cortex and stops when the pulse stops; there is no radiation, no anesthesia and no induced seizure. Cognitive testing in the registration trial found no change at four or six weeks, and a review of 31 randomized trials found no reliable cognitive side effects. What can be said honestly is that no permanent adverse effect has been demonstrated — and that follow-up in this literature runs to about a year.
For a lot of people the first few sessions do. Pain at the treatment site was reported by 35.8% of people on active stimulation against 3.8% on sham, so it is a real effect of the stimulation rather than nerves or expectation. It is usually described as tapping or a sharp pressure, it commonly eases as you get used to it, and the operator can move the coil and ramp the intensity. Say something rather than enduring it.
Metal that cannot be removed within 30 cm of the coil is a contraindication: aneurysm clips, stents, implanted electrodes, bullet fragments. So are implanted stimulators including deep brain stimulators, cochlear implants and vagus nerve stimulators, and their leads. Standard fillings, single-post dental implants and bridge work are explicitly fine. A pacemaker or a defibrillator is not an automatic no, but it requires caution and a conversation. None of this is for you to judge — tell the clinic everything and let them decide.
Nobody knows, and the label says that rather than saying no. Pregnant and nursing patients were excluded from the trials, so safety and effectiveness were never established in that group. The safety guidelines list pregnancy under conditions of increased or uncertain risk, not under contraindications. TMS is used in pregnancy in practice, off-label, and that is a risk-benefit judgment for an obstetrician and a psychiatrist together. It is not a question this page can answer for you.
There is no head-to-head trial, so anyone who gives you a straight yes is guessing. What can be said is that the risk profiles are shaped differently. TMS has no systemic exposure — nothing circulates, so there is no weight gain, no sexual side effects, no discontinuation syndrome and no interaction with your other medications. Its serious risk is rare, immediate, and happens in a supervised room. Medication risks are commoner, milder and spread across months at home. Both carry the same bolded warning about worsening depression and emergent suicidal thinking, because that warning is about the illness.
Sources
Read next
Pillar
What is TMS therapy?
/tms-therapy
Spoke
TMS therapy side effects
/resources/tms-side-effects
Spoke
Why some people say TMS ruined their life
/tms-therapy/why-people-say-it-ruined-their-life
Listings show each clinic's phone number, website and hours. The questions above are the ones worth asking before you book: which device and coil they use, what your screening will cover, and who is in the room during a session.
Browse all locations →