How long does it take to form a habit? What the 66 days measured, and where 21 came from

How long it takes, study by study

The studies that have measured it put the typical time at about two months, with individuals ranging from under three weeks to the better part of a year. Every figure below is from the paper itself or its published abstract, and each comes with the way it was measured, because "formed" means something different in each.

Two things stand out. No study found 21 days as a typical figure, and the self-report studies watched people for 84 or 90 days, so every estimate longer than that is a projection of a fitted curve, not a day anyone reached while being observed.

What Lally's 66 days actually measured

The 66 days is the median time for a person's sense of automaticity to reach 95% of its own ceiling, estimated from a curve fitted to 84 days of daily ratings. It is not the day a habit "switched on", and it is not a figure observed in everyone.

In the study, published in the European Journal of Social Psychology (volume 40, pages 998 to 1009; online in 2009, in print in 2010), Phillippa Lally, Cornelia van Jaarsveld, Henry Potts and Jane Wardle at University College London asked 96 volunteers to choose an eating, drinking or activity behaviour and do it once a day in the same context, "for example 'after breakfast'", for 12 weeks. Each day they recorded whether they did it and completed the Self-Report Habit Index, a questionnaire of agree-or-disagree statements about a behaviour. Automaticity was calculated from a subset of the index's statements.

The researchers then fitted a curve to each person's scores. Automaticity rose fastest in the first days and weeks and then flattened towards a ceiling, the asymptote. The time to habit was defined as the day the curve reached 95% of that person's ceiling. Because the curve flattens, the last 5% would take a very long time, so the 95% point is a convention for "close enough to the top".

The time estimates come from the fitted curves, so they describe the people whose data the model could fit, not all 96. The abstract also reports that people who did the behaviour more consistently had better-fitting curves. That is the detail most retellings drop: for more than half of the 82 people who gave enough data, the study could not produce a reliable time at all.

Each person's ceiling was their own. Someone whose automaticity levelled off at a modest score reached 95% of it as surely as someone who levelled off near the top, so a short time does not mean a stronger habit, only an earlier plateau.

Median, mean and the 18-to-254 range

66 days is a median: half the estimates were shorter and half longer. The 2024 systematic review tabulates it that way and lists no mean for the study. The confusion is understandable, because Lally's own group described it as "an average of 66 days" in a 2012 article for family doctors in the British Journal of General Practice, which suggested telling patients to expect "around 10 weeks".

The distinction matters with a range this lopsided. Estimates ran from 18 days to 254, and long times like 254 pull a mean upwards while leaving the median where it is. A planner who reads "average 66" as "most people are done by about day 66" is reading it about right; one who reads it as "everyone is done by day 66" will be wrong for half the people in the study.

Three other numbers are often mixed into this one:

• 335 days is not Lally's. It is the longest estimate in Keller and colleagues' 2021 trial, and the 2024 review's "4 to 335 days" is the range across studies.
• 106 to 154 days are the 2024 review's "means", but they come from a single study, Fournier and colleagues' 2017 stretching trial, with 106 days for the morning group and 154 for the evening group. They are not means of the same data the 59-to-66-day medians come from.
• 254 days is longer than the 84 days of observation. It is where one person's fitted curve was projected to reach 95% of its ceiling, not a day that person was seen to reach.

Where "21 days" comes from

The 21 days traces to Maxwell Maltz, a plastic surgeon, in his 1960 self-help book Psycho-Cybernetics, and he was writing about self-image, not habits. In the first edition (Prentice-Hall, 1960), whose text can be searched through Open Library, Maltz wrote that "it usually requires a minimum of about 21 days to effect any perceptible change in a mental image".

His evidence was what he had seen in practice: following plastic surgery "it takes about 21 days for the average patient to get used to his new face", and after an amputation "the 'phantom limb' persists for about 21 days". From such observations he concluded that "it requires a minimum of about 21 days for an old mental image to dissolve and a new one to jell". A heading in the preface reads "Reserve judgment for 21 days".

Three words changed on the way to the popular version. "Minimum" became an exact figure, "about" was dropped, and "mental image" became "habit". The 2012 article by Gardner, Lally and Wardle, citing the same book, says the 21-day claim "appears to have originated from anecdotal evidence". No study in this guide found a typical time anywhere near it except hospital handwashing, a behaviour repeated many times a shift.

What newer research adds

Since 2010, the studies agree that habits form along a curve and that the time varies widely, but there are still only a handful that measure time at all. The 2024 review is the most cited summary, and its headline numbers need their context.

The 2024 systematic review

Ben Singh, Andrew Murphy, Carol Maher and Ashleigh Smith at the University of South Australia reviewed the evidence in Healthcare (volume 12, issue 23, article 2488, 9 December 2024). They included 20 studies with 2,601 participants, and rated 11 of them at high risk of bias. Only four studies reported a time to habit formation, which is where the 59-to-66-day medians and the 4-to-335-day range come from.

The review's meta-analysis, a standardised mean difference of 0.69 across 12 studies, measures how much habit scores rose after an intervention, not how long it took. Its findings on what helps were that morning practice and habits people chose for themselves showed greater strength. The authors name the small number of studies as the main limitation.

Morning against evening

Marion Fournier and colleagues' 2017 trial in Health Psychology assigned 48 students to do one stretch either on waking or before bed for 90 days. Extrapolated curves put automaticity at 105.95 days for the morning group and 154.01 days for the evening group, and the authors linked the difference to the daily rhythm of the hormone cortisol.

Gym against handwashing

Anastasia Buyalskaya and colleagues' 2023 study in PNAS used no questionnaire at all. It measured how predictable behaviour became from its context, in records of 30,110 gym members and 3,124 hospital workers. Hospital handwashing became predictable in "on the order of a couple of weeks", about 9 to 10 shifts. For the gym the original paper said 4 to 7 months, but a correction published in August 2023 changed it to 68 to 78 days, about 2 to 3 months. Copies of the original figure still circulate.

Agreeing on how to measure it

In 2022 Benjamin Gardner, Amanda Rebar and Phillippa Lally set out fourteen criteria for studies that track habit formation in Cogent Psychology and found only five existing studies that met them. In a 2024 review the same group, with Sanne de Wit, went further: habit formation "may be neither necessary nor sufficient to sustain real-world behaviour change". A habit helps a behaviour survive a dip in motivation; it does not guarantee the behaviour continues.

What the studies cannot tell you

The 66 days is one figure from one small study of volunteers doing simple things, and it describes how automatic a behaviour felt, not whether it lasted. Six limits apply to every number above:

1. Self-report. Lally, Keller and Fournier asked people how automatic the behaviour felt. Only the 2023 study measured behaviour directly, and it measured predictability instead of automaticity.
2. Simple, chosen behaviours. A daily piece of fruit or a single stretch is not a training plan or a new diet. The 2024 review found most of its evidence on flossing and diet.
3. Short observation. At 84 or 90 days, every longer estimate is a projection of the curve.
4. Missing fits. Lally's curve fitted well for 39 of the 82 people who gave enough data. The study cannot say how long the others took.
5. Days, not repetitions. The 2022 criteria paper notes that the tracking studies analysed time rather than the number of repetitions, although repetition is what builds a habit. A behaviour done twice a day may form on a different clock from one done once.
6. Forming, not breaking. None of these studies timed how long it takes to stop an old habit.

Lally's study also had too few people per behaviour type to compare them: the 2022 paper notes it was underpowered to detect such differences, and the 2024 review reports no significant difference between its exercise, water and fruit habits.

Does missing a day reset a habit?

No. In Lally's study, missing one opportunity to perform the behaviour "did not materially affect the habit formation process". The 2022 criteria paper reports that the two studies which tested missed days found a temporary effect that faded, and the 2012 summary says automaticity gains "soon resumed after one missed performance".

A missed day does reset a streak, because a streak counts consecutive days, not habit strength. That is why the never-miss-twice rule in the streak guide works better than "never miss": a single miss costs the number, not the habit. The research supports single misses being harmless; it does not say how many misses in a row start to matter.

A worked example: tracking a habit with a counter and a streak

Track two numbers and one rating: a streak for motivation, a running total of days done for the repetitions, and a weekly automaticity score to see your own curve flatten. The numbers below are an illustration, not data from a study.

The habit is ten press-ups after brushing teeth in the morning, a behaviour with a fixed cue, as in Lally's study.

1. In the habit counter, add "Press-ups after teeth" with a daily target of
2. Leave the day boundary at 04:00 unless your mornings start later.
3. In the online counter, add "Press-up days, total". Tap it once on each day the target is met, and never reset it. This is the repetition count the streak loses when it breaks.
4. Every Sunday, rate four statements from the Self-Report Behavioural Automaticity Index, the four-item automaticity scale derived from the habit index in 2012, on a 1-to-7 scale of agreement: doing press-ups after brushing my teeth is something "I do automatically", "I do without having to consciously remember", "I do without thinking" and "I start doing before I realise I'm doing it". Add them up, for a score from 4 to 28, and write it down: the habit counter keeps today's count and the streaks, but no dated history.

Reading it takes three steps:

1. The score rose 4, 3 and 2 points between weeks 1 and 4, then by 2 or fewer a week, and stopped at 25 from week 10. Fast early gains slowing to a plateau is the shape Lally's curve describes.
2. The streak and the total tell different stories. In week 9, after two missed days in a row, the streak reads 3 and the best streak still reads 26 (days 11 to 36). The total reads 59: the repetitions were not lost, and the score dipped by one point and recovered the following week.
3. The plateau, not a date, is the finish. This example flattened at 25 from week 10, around day 70, close to Lally's median by coincidence. A person whose score was still climbing at week 12 would simply be further along a longer curve, as many of Lally's volunteers were projected to be.

Once the weekly score has stopped rising and the press-ups happen before you think about tapping, the streak guide's advice on when to stop counting applies: the counter has done its job.

What this guide does not cover

This guide covers the published research on how long a new, simple, repeated behaviour takes to feel automatic, as of 4 November 2026. It does not cover breaking habits, addiction or compulsive behaviour, for which the craving tracker records urges and points to professional help; nor does it give medical or exercise advice. The numbers describe groups of volunteers, so they cannot predict one person's time.

Simple Online Counter is a free, browser-based set of counters. Its habit counter counts a daily quantity against a target and keeps the current and best streak, with the day ending at an hour you choose, and nothing leaves the browser.