How long does it take to form a habit? Read the curve, not the deadline

The spread between people is wider than the average itself, which makes “how many days?” a slightly wrong question. The studies that track habit formation day by day keep finding the same pattern: big gains early, smaller gains later, then a plateau. Where you are on that curve tells you more than how many days have passed since you started.

Where the 21-day rule came from

The 21-day figure goes back to Maxwell Maltz, a plastic surgeon who wrote the 1960 self-help book Psycho-Cybernetics. In the preface he noted that patients needed about three weeks to get used to a new face, and that people felt a phantom limb for about as long after an amputation. His conclusion was about self-image, and he hedged it (Gardner & Meisel, 2012):

It usually requires a minimum of about 21 days to effect any perceptible change in a mental image.

Maxwell Maltz, Psycho-Cybernetics (1960)

Two things went missing on the way to the self-help shelf. Maltz wrote “a minimum of about”, and he was describing getting used to something, which psychologists call habituation. Learning to do something without thinking is a different process. The UCL researchers who traced the myth think the two ideas simply merged as the claim was repeated.

What the 66-day study measured

The number most people quote now comes from Lally et al., 2010, published in the European Journal of Social Psychology. 96 volunteers each chose one eating, drinking or activity behaviour and did it once a day in the same context, for example “after breakfast”, for 12 weeks. Every day they recorded whether they had done it and answered a short questionnaire about how automatic it felt.

82 people gave enough data to analyse. The researchers fitted a curve to each person’s automaticity scores over the 84 days. The model fitted for 62 people and fitted well for 39. For that group, the time to reach 95% of their own plateau had a median of 66 days. The fastest person got there in 18 days. The slowest hadn’t plateaued when the study ended and was forecast to need about 254 (Gardner & Meisel, 2012).

Another finding from the same study gets less attention than it deserves: missing one opportunity to perform the behaviour did not materially affect the habit formation process. A single slip doesn’t send you back to day one. There is more on that in what a missed day does to a streak.

What “automatic” means in this research

Here a habit is a behaviour that a familiar context sets off by itself. Gardner et al., 2012 define habits as behaviour patterns, learned in a context, that are “elicited automatically upon encountering associated contexts”. Automaticity is the part you can feel: the behaviour starts before you have decided to do it.

Researchers measure it by asking. The Self-Report Habit Index has 12 items, and Lally’s team used the ones about automaticity. Gardner and colleagues later boiled those down to four statements, the Self-Report Behavioural Automaticity Index. For a behaviour X, you rate how much you agree that it is something…

• I do automatically
• I do without having to consciously remember
• I do without thinking
• I start doing before I realise I’m doing it

Two people can both “form a habit” and end up in different places. In Lally’s data, some people’s automaticity levelled off below the halfway point of the 42-point scale, while others reached the very top (Gardner & Meisel, 2012). A plateau tells you the gains have stopped. It doesn’t tell you the habit is strong.

The habit curve, week by week

The curve is the most practical thing to take from this research. Lally’s team modelled automaticity as an asymptotic curve: each repetition adds a bit less than the one before. If you draw that curve with the median numbers, starting from zero and reaching 95% of the plateau on day 66, the arithmetic looks like this:

• End of week 1: about 27% of the way to your plateau
• End of week 2: about 47%
• Day 21: about 61%
• End of week 4: about 72%
• End of week 6: about 85%
• Day 66: 95%, after which each extra week adds a point or two

The weekly gains tell the story better than the totals: roughly 27 points in the first week, 20 in the second, 14 in the third, 11 in the fourth, and by week eight about 3. Real ratings bounce around more than a smooth line does, and your own curve might be steeper or flatter. The shape still explains a common experience. The first fortnight feels like obvious progress, and weeks five to nine feel like nothing is happening, even though you are still gaining ground.

This is probably also why the 21-day claim survives. On the median curve, three weeks is when the change is easiest to notice, because you have just come through the steepest stretch. About 40% of the climb is still ahead.

Why some habits take longer than others

Complexity is the obvious suspect. In Lally’s study, simple behaviours such as drinking a glass of water peaked sooner than more involved ones such as doing 50 sit-ups (Gardner, Lally & Wardle, 2012). The difference between the eating, drinking and exercise groups was not statistically significant, though, and with 39 well-fitting curves the study was too small to settle it (Singh et al., 2024).

Consistency matters too. In Lally’s data, people who did the behaviour more consistently produced curves that fitted better. A 2021 trial by Keller and colleagues with 192 adults found that repeatedly carrying out the plan was a key predictor of automaticity. Linking a nutrition behaviour to a daily routine worked about as well as linking it to a clock time. If-then plans are one way to make that link explicit.

Time of day may play a part. In a study of 48 students who did a stretch either on waking or before bed, extrapolated curves put automaticity at about 106 days for the morning group and 154 for the evening group, and the authors linked the gap to the daily rhythm of cortisol. It is one small study and it can’t tell you which time suits you, but it fits a wider pattern covered in morning versus evening habits.

How long does it take to form a habit across studies?

The 2024 systematic review by Singh et al. in the journal Healthcare pulled together 20 studies with 2,601 participants, covering behaviours such as physical activity, diet, drinking water and flossing. Habit scores rose clearly from before to after the interventions. Only four of the studies reported how long formation took: medians of 59 to 66 days, means of 106 to 154 days, and individual times from 4 to 335 days.

The gap between medians and means is informative. A few slow people pull the average up, so a typical person is quicker than the mean, though still far slower than 21 days. The authors’ practical advice is to expect “at least two to five months” for automaticity. They also found that simple, repetitive behaviours with clear cues, such as flossing or drinking water, showed larger effects than healthy eating or exercise, and noted that this might reflect the interventions rather than the behaviours.

What the research doesn’t tell you

• The samples are small. The 66-day median rests on 39 people whose curves fitted well, out of 96 who started.
• It is all self-report. Automaticity is how automatic a behaviour feels to the person doing it, and rating it every day may itself change how people experience it.
• Long times are forecasts. Lally and Keller followed people for 84 days and Fournier for 90, so values like 254 or 335 days come from extending curves past the data.
• Not everyone gets there. In Keller’s trial, only 23% of participants reached the researchers’ threshold for habit formation (Singh et al., 2024).
• These are health behaviours people chose to start. The review excluded studies about stopping a behaviour, such as smoking, so none of these numbers say how long it takes to break a habit.
• Quality is uneven. The review rated 11 of its 20 studies as having a high risk of bias.

How to tell where you are on your curve

You can’t know your number in advance, but you can watch your own curve take shape. A simple routine that follows the research:

1. Tie the behaviour to one stable cue: a place, a time, or something you already do every day. Use the same cue every time.
2. Once a week, rate the four statements above on a scale you keep fixed, say 1 (disagree) to 7 (agree), and note the average.
3. Compare week with week. Early on the score should jump; later it should creep. When it stops moving for two or three weeks, you have probably reached your plateau.
4. If the plateau is low, make the behaviour simpler or the cue more consistent before deciding you just need more time. Simple behaviours tied to a cue tended to become automatic sooner in these studies.
5. If you miss a day, do it the next day. In Lally’s data one missed opportunity made no material difference.

Plan for a couple of months and expect the second month to feel slower than the first. If you want to know whether tracking itself helps, the evidence is in does habit tracking work?