# How Long Does It Take to Form a Habit? What the Research Found

> The 21-day rule comes from a 1960 self-help book. Here is what UCL's 66-day study and later data show, plus a plan with check-ins at weeks 2, 6, and 10.

- Canonical page: https://learnhubplay.nl/blog/how-long-to-form-a-habit
- Published: 2026-10-09
- By: LearnHubPlay, an independent software studio in Amsterdam
- Topic: Habits & Routines
- Tags: Habit formation, Habit tracking, Behaviour change, Research explainers

You start a daily habit on a Monday, you have read that it takes 21 days, and on day 21 it still takes effort. The usual verdict is that you lack discipline. The research offers a plainer reading: the 21-day figure was never measured. The best-known measurement, a 2010 UCL study, found a median of 66 days among the 39 of 82 participants whose scores followed a habit curve, with individuals ranging from 18 days to a projected 254.

**The short answer.** For a simple behaviour tied to a once-a-day cue, plan on about two to three months of near-daily repetition. The median was 59 to 66 days among the people whose habits formed in two 84-day studies, and a 2024 review of 20 studies advises allowing at least two to five months. Exercise ran slower in the UCL data (a projected median of 91 days). By the end you will be able to test any "habit in X days" claim and judge your own progress with ratings at weeks 2, 6, and 10, instead of borrowing a deadline.

## Where the 21-day number came from: a surgeon's anecdotes about self-image

Maxwell Maltz was a plastic surgeon, and the 21 days comes from the preface of his book *Psycho-Cybernetics* (1960). Gardner and Meisel's UCL post quotes the line (pp. xiii–xiv): "It usually requires a minimum of about 21 days to effect any perceptible change in a mental image."

His evidence is three observations: patients take about 21 days to get used to a new face after plastic surgery, a phantom limb persists about 21 days after an amputation, and a new house takes about three weeks to feel like home.

So the claim is about mental images, not habits; it is a minimum, not an average; and it rests on anecdotes. Gardner and Meisel [suggested in 2012](https://blogs.ucl.ac.uk/bsh/2012/06/29) that the difference between getting used to something and forming a habit "was lost in translation somewhere along the line." A 2024 systematic review credits the idea to Maltz's book and says its findings "clearly refute the popular notion that habits can be formed in approximately 21 days."

## What Lally's team had 96 people do for 84 days

Phillippa Lally, Cornelia van Jaarsveld, Henry Potts, and Jane Wardle at University College London published their study in the *European Journal of Social Psychology* (online 2009, in print 2010). The design explains what the number means:

- **Participants.** 96 volunteers (66 women, 30 men), mostly postgraduate students, mean age 27.
- **The task.** Each picked one healthy eating, drinking, or exercise behaviour they did not already do and could attach to a salient daily event, such as eating fruit with lunch or running for 15 minutes before dinner. Of the 96, 27 chose eating, 31 drinking, 34 exercise, and 4 something else, such as meditation. They tried to do it daily for 84 days.
- **The measure.** Each day they reported whether they had done it and rated seven statements from a habit questionnaire, such as "I do automatically" and "I do without thinking." Scores ran from 0 to 42. This measures automaticity (how little thought the behaviour needs), not days completed.
- **The analysis.** The team fitted a rising curve to each person's scores, *y = a − b·e^(−cx)*, which levels off at a ceiling. Here *a* is the ceiling, *b* is the climb from the starting score to the ceiling, and *c* is how fast the climb goes. A curve never quite touches its ceiling, so "formed" meant the day the fitted curve reached 95% of it.

Nobody was rewarded for doing the behaviour (the £30 study payment did not depend on it), and the cue had to be a salient event that happened once a day, such as lunch or dinner.

## How to read the 66-day median: 39 good curves out of 82 people

The headline number describes a subgroup. 96 people started and 14 did not enter data after day 60, leaving 82. A curve could be fitted for 62, and 39 of those were good fits (R² above 0.7 and a fitted ceiling between 21 and 49), which is 48% of the 82. R² is the share of the ups and downs in a person's scores that the curve tracks. For those 39, the median time to 95% of the ceiling was 66 days, the middle half ran from 39 to 102, and the full range was 18 to 254.

Table 1 of the paper shows who sat in each group and how often they did the behaviour:

| Group of 82 | People | Median share of days done |
|---|---|---|
| Good fit | 39 | 88% |
| Ceiling above 49 (implausibly high) | 3 | 82% |
| Model could not be fitted | 12 | 80% |
| Flat curve (no rise in scores) | 8 | 63% |
| Ceiling under 21 (rose, but never to habit level) | 4 | 61% |
| Poor fit (R² under 0.7) | 16 | 60% |

**66 days describes the people whose curves flattened.** The authors' summary is that approximately half did not perform the behaviour consistently enough to reach habit status. The table is more specific: the bottom three groups, 28 people, did it on about 60% of days, while the other 15 did it on about 80% or more, and the authors read them as slow formers who "would have reached a plateau if the recording had continued for longer." So 66 days probably understates the time for slow but consistent people. The good-fit group's median of 88% of days (middle half: 80% to 94%) is a correlation, but it is the clearest consistency target the data offer.

**Anything beyond day 84 is a projection.** The study ended at 84 days, so longer times, including the 254-day extreme and the 91-day exercise median, were extrapolated.

**The measure is self-report.** It captures felt automaticity and cannot separate a true habit from a flexible, goal-driven routine that merely feels easy. The authors also say it is hard to know whether answering the same questions every day changed people's answers. Keller's trial below is the closest later check, with a different design.

Asked in a [Surrey news interview](https://www.surrey.ac.uk/news/does-it-really-take-66-days-form-habit-we-asked-expert-dr-pippa-lally) (January 2026) whether it really takes 66 days, Lally answered: "for most people most of the time, no." What her study showed, she said, is that "how long habit formation takes is highly variable."

### A typical curve in numbers: where day 21 sits

The paper's Table 2 gives median parameters for the 39 good fits: ceiling *a* = 35, climb *b* = 30 (so a start near 5), rate *c* = 0.042. Plugged into the formula, they give this illustrative curve. It is not any real person, and combining median parameters does not give the median outcome.

| Day | Score (0–42) | Share of ceiling |
|---|---|---|
| 0 | 5.0 | 14% |
| 7 | 12.6 | 36% |
| 21 | 22.6 | 65% |
| 42 | 29.9 | 85% |
| 66 | 33.1 | 95% |
| 84 | 34.1 | 97% |

By day 21 the curve has covered about 59% of the climb and crossed 21, the paper's neutral mark (an average answer of "neither agree nor disagree"), around day 18. Early repetitions add the most: 7.6 points in week one against about 1.3 in week seven. Hold *a* and *b* fixed and move only the rate between its quartiles (0.028 and 0.064), and the 95% mark lands at roughly 101 and 44 days.

<figure><img src="/images/blog/how-long-to-form-a-habit-curves.svg" alt="Three rising curves on a 0 to 42 scale over 120 days, all starting near 5 and flattening toward a ceiling of 35. The fast curve reaches 95 percent of its ceiling at about day 44, the median-rate curve at about day 68, and the slow curve at about day 101. A vertical line marks day 21, where the median-rate curve is at 22.6. Days after 84 are shaded as projection." width="960" height="600" /><figcaption>Illustrative curves built from the median ceiling and climb and the quartile rates in Lally et al. (2010), Table 2. The median-rate curve reaches 95% at day 68 against the published median of 66, because combining medians does not give the median. Not any real person.</figcaption></figure>

## Three later sources: two months or more for most behaviours, much less for a very frequent one

| Study | Who and how | Result | Caveat |
|---|---|---|---|
| Keller et al. 2021, *British Journal of Health Psychology* | 192 adults (mean age about 25, 87% female) planned a nutrition behaviour around a routine or clock-time cue; 84 days | 117 followed the protocol (half chose drinking water). 95 (81%) showed rising automaticity; 27 (23%) reached a plateau above the scale mean, after a median of 59 days (range 4–335) | Self-reported; young, mostly female sample; no control group, so the daily questionnaire itself may have helped |
| Buyalskaya et al. 2023, *PNAS* | Objective records of 30,110 gym members and 3,124 hospital workers | Gym: about 68–78 days (roughly 2–3 months) to reach 95% of the fitted curve, per the authors' 2023 correction (the original printed 122–226). Handwashing: 9 to 10 shifts, about 220 washing opportunities | "Habit" meant predictability, not felt automaticity; gym members with under a year of data excluded |
| Singh et al. 2024, *Healthcare* | Review and meta-analysis of 20 studies, 2,601 participants | Four studies reported time: medians 59–66 days, means roughly 91–154 | 11 of 20 at high risk of bias; three of the four timing studies (Lally 2008, Lally 2010, Keller 2021) involve Lally, so they are not independent |

The review's authors advise expecting at least two to five months, while its abstract says habits "can start forming within about two months," so treat two months as the early end.

The gym and handwashing gap hints at why numbers differ: a hospital worker met the cue about 220 times in nine or ten shifts, while the median gap between gym visits was about a week. The authors suggest frequency and complexity as one possible explanation, but did not test it as a cause.

## How long to plan for, and what seems to speed things up

Lally and Keller both report times by behaviour, which gives the closest thing to a planning table:

| Behaviour | Source | Median days to 95% of ceiling | Spread |
|---|---|---|---|
| Eating, such as fruit with lunch | Lally, 10 people | 65 | middle half 35–106 |
| Drinking, mostly water | Lally, 15 people | 59 | middle half 39–75 |
| Exercise, such as a 15-minute run before dinner | Lally, 13 people | 91 (projected past day 84) | middle half 44–118 |
| Nutrition behaviours, half of them drinking water | Keller, 27 of 117 people | 59 | range 4–335 |

Lally's three groups did not differ significantly (p = 0.328) and the study was not powered for subgroup tests, so read the table as a hint. As a rule of thumb, not a finding: judge a simple daily behaviour at about week 10, allow into month three or four for exercise, and do not apply these numbers to anything you will do less than daily. No source here times writing, studying, or coding habits; for those, make the action the unit, not the output ("open the draft and write for 10 minutes", not "write a chapter").

What seems to help:

- **A cue you meet daily in the same context.** Lally's participants had to pick a once-a-day event, and Wood and Rünger's 2016 review in the *Annual Review of Psychology* names repetition in stable contexts as the foundation of habit formation. In Keller's trial, repeated plan enactment was a key predictor of automaticity, and routine and clock-time cues worked equally well (14 versus 13 successful habit-formers), so choose whichever you will actually meet daily.
- **A behaviour that feels rewarding to you.** In Keller's trial, habit-formers reported higher intrinsic reward at baseline, and intrinsic reward was a unique predictor of who formed one. Singh's review reported self-selected habits as generally stronger, though only two included studies tested this. Lally's participants got no rewards for the behaviour itself, but they had chosen behaviours they wanted.
- **Keeping it simple.** Lally's authors read the exercise group taking about one and a half times longer as support for the idea that complexity affects how fast automaticity develops, though the difference was not significant.

The honest summary: nothing here shows a lever that cuts the median dramatically. The best-supported advice is dull and useful: repeat it almost daily, keep it small, and tie it to a cue that reliably happens.

## A habit formation plan with check-ins at weeks 2, 6, and 10

This plan follows the advice Lally's group gave clinicians in a [2012 *British Journal of General Practice* article](https://pmc.ncbi.nlm.nih.gov/articles/PMC3505409/): pick a simple daily action, plan when and where, do it each time the cue comes up, expect it to feel automatic within about 10 weeks, and rate how automatic it feels each week. A when-and-where plan like this is what psychologists call an implementation intention (Keller's team cites Gollwitzer's 1999 work on it). Fill this in once:

```text
HABIT FORMATION PLAN
Start: ______   Check-ins: day 14 ___   day 42 ___   day 70 ___

1. Cue (one event, same place, once a day):
   After I ____________________ in/at ____________________,
2. Full version (normal day): I will ____________________.
3. Tiny version (bad day, under 2 minutes, still counts):
   I will ____________________.
4. Density target: at least 12 of every 14 days, tiny version included.
5. Backup if the cue did not happen: I will do it at ________.
6. Weekly rating, 1-7 (1 = strongly disagree, 7 = strongly agree):
   "[My behaviour] is something I...
   a. do automatically
   b. do without having to consciously remember
   c. do without thinking
   d. start doing before I realize I'm doing it"
   Average of a-d:   Wk1 ___   Wk2 ___   Wk3 ___ ...
```

The four statements are the Self-Report Behavioural Automaticity Index, developed by Gardner, Abraham, Lally, and de Bruijn in 2012. The 12-of-14 target comes from the 88% median among Lally's participants whose curves plateaued. It is a benchmark, not a guarantee.

**A worked example (hypothetical).** Say you want a ten-minute walk after dinner. Cue: after I put the plates in the dishwasher. Tiny version: shoes on, two minutes outside. Target: 12 of 14 days, or 72 of the first 84. Exercise ran slowest in Lally's data, though not significantly, so budget three months.

The cut-offs below are our rule of thumb, informed by the gap between the plateau group's 88% consistency and the poor-fit groups' roughly 60%. Compare the shape of your ratings (rising, then flat), not the numbers, with the curve table above: the plan's 1–7 average is a different instrument from Lally's 0–42 score. A rating of 4 is the scale midpoint, a convention rather than a validated cut-off: Lally's team said scores above neutral are not necessarily habits.

| Check-in | Look at | Then |
|---|---|---|
| Day 14 | Days done of the last 14 | 12–14: change nothing. 9–11: the cue is leaky, so tie it to a daily event or use the backup. 8 or fewer: the plan is too big, so shrink it instead of pushing harder |
| Day 42 | Density, and rating versus week 1 | Density 11+ and rating rising: normal (a typical curve is at about 85% of its ceiling, and a quarter of Lally's habit-formers did it on 80% of days or fewer, about 11 of 14). Density 11+ but rating flat below 4: split the behaviour or move the cue. Density under 11: fix density first |
| Day 70 | Last three weekly ratings | Still rising: keep going. Flat at 4 or above: likely your plateau, so check density monthly. Flat below 4: still effortful, so treat it as a deliberate routine and keep tracking |

Other guides on this site use looser density defaults, such as 10 of 14 for a daily habit. This plan uses 12 of 14 because Lally's plateau group had a median of 88%.

> **Tip:** Add the four-statement rating to your [weekly review](https://learnhubplay.nl/blog/weekly-review-template). Count days on a paper calendar or in a spreadsheet, and read the result as "12 of the last 14 days" instead of a streak. A counter that resets to zero after one miss contradicts Lally's 0.29-point dip for a missed day; for why streaks backfire, see [why most habit trackers get abandoned](https://learnhubplay.nl/blog/habit-tracking-that-sticks).

> **Warning:** Do not treat 66 days as a deadline. It was the median for people whose curves flattened, and a rating still climbing at day 66 is inside the normal range.

### If your rating will not rise

Our rules of thumb, each tied to a source:

| Symptom | Likely cause | What to try |
|---|---|---|
| Density is fine but the rating is flat | The action varies day to day; Gardner, Lally, and Wardle say variation is incompatible with automaticity | Fix one exact action in one place |
| The cue is a phone alert | In one study (Stawarz et al., cited by Wood and Rünger), reminders raised repetition but impeded automaticity | Move the cue to an event that already happens, such as finishing a meal |
| You do it but do not want to | Intrinsic reward predicted who formed a habit in Keller's trial | Redesign it into a version you want |
| It feels easy but never automatic | Self-report cannot separate a habit from a goal-driven routine | Keep tracking density and treat it as a reliable routine |

## Where to start this week

1. Choose one behaviour and fill in items 1 to 4 tonight, with a tiny version you would do on your worst day this month.
2. Put the day 14, 42, and 70 check-ins in your calendar, and judge the plan at day 70, not day 21.
3. When you meet the next "habit in X days" claim, ask four things: who was studied and for how long, what was measured (felt automaticity, objective predictability, or days completed), what share of participants reached a plateau, and whether the time was observed or projected past the study window.

## Frequently asked questions

### Does missing a day reset the clock?

Not in Lally's data. Across 140 single misses (each right after three days of doing the behaviour, in 55 people), automaticity fell by an average of 0.29 points on the 42-point scale. Over two days the gain was 0.55 points with a miss in between against 0.79 without one, and the authors concluded that a miss "did not materially affect the habit formation process." A week-long lapse is different: Armitage (2005), discussed in the paper, found it hurt later performance. As a rule of thumb, not a finding, do the tiny version at the next cue and skip catch-up sessions.

### Can I build several habits at the same time?

The sources here do not say; each of Lally's participants worked on one behaviour. As a rule of thumb, not a finding, give every habit its own cue and tiny version, and add the next only after the first has met its density target for several weeks.

### How do I know when a habit counts as formed?

There is no official line. Researchers usually define it as the plateau of felt automaticity, which Lally's team set at 95% of each person's fitted ceiling. Our rule of thumb, not a validated cut-off, is a weekly rating of 4 or above that has stopped climbing for three weeks.

### What if my routine changes because of a trip, illness, or a new job?

Wood and Rünger note that life transitions, such as switching jobs or moving house, change the cues that trigger habits. The sources here do not test what that does to a half-formed habit, so this is a rule of thumb: restate the cue for the new setting, run the tiny version for a week, and count density from the restart instead of resetting the 70-day plan.

### Does this apply to breaking a habit?

Not directly. Lally's participants all chose behaviours they did not already do, and Singh's review excluded studies aimed at stopping behaviours. Wood and Rünger point to changing the cues around you, and making the old behaviour harder to reach, as in [this guide to adding friction](https://learnhubplay.nl/blog/blocking-distractions-with-friction), is one practical way to do that, though the sources here did not test it. Gardner, Lally, and Wardle advise forming a new behaviour rather than stopping an old one.

## Sources and further reading

- [How are habits formed: Modelling habit formation in the real world](https://repositorio.ispa.pt/handle/10400.12/3364) — Lally, van Jaarsveld, Potts & Wardle, European Journal of Social Psychology, 2010 (DOI 10.1002/ejsp.674). Design, group table, 66-day median, range, missed days.
- [Time to Form a Habit: A Systematic Review and Meta-Analysis](https://pmc.ncbi.nlm.nih.gov/articles/PMC11641623/) — Singh, Murphy, Maher & Smith, Healthcare, 2024. Four timing studies, bias limits, the 21-day conclusion.
- [Habit formation following routine-based versus time-based cue planning](https://minerva-access.unimelb.edu.au/server/api/core/bitstreams/85b44c09-6a67-58ce-ac9d-8d12c3e421bb/content) — Keller et al., British Journal of Health Psychology, 2021 (DOI 10.1111/bjhp.12504). The 192-person trial.
- [What can machine learning teach us about habit formation?](https://pmc.ncbi.nlm.nih.gov/articles/PMC10151500/) — Buyalskaya et al., PNAS, 2023, with the authors' [correction](https://pmc.ncbi.nlm.nih.gov/articles/PMC10450439/) (the uncorrected text gives 122–226 days for the gym). Gym and handwashing timing from objective records.
- [Psychology of Habit](https://dornsife.usc.edu/wendy-wood/wp-content/uploads/sites/183/2023/10/wood.runger.2016.pdf) — Wood & Rünger, Annual Review of Psychology, 2016 (pre-publication copy, DOI 10.1146/annurev-psych-122414-033417). Stable contexts, reminders, old habits, life transitions.
- [Making health habitual](https://pmc.ncbi.nlm.nih.gov/articles/PMC3505409/) — Gardner, Lally & Wardle, British Journal of General Practice, 2012. Practical steps, the 10-week expectation.
- [Towards parsimony in habit measurement](https://pmc.ncbi.nlm.nih.gov/articles/PMC3552971/) — Gardner, Abraham, Lally & de Bruijn, Int. J. Behav. Nutr. Phys. Act., 2012. The four-item rating.
- [Busting the 21 days habit formation myth](https://blogs.ucl.ac.uk/bsh/2012/06/29) — Gardner & Meisel, UCL, 2012. The Maltz passage and the habituation explanation.