AgeDart › Blog

How good are we at guessing someone's age? What the studies say

Published October 16, 2026 · AgeDart

We guess ages all day: a new colleague, someone on the train, a face on a dating app. It feels easy. Research says we're decent but far from precise, and that our mistakes follow predictable patterns. Knowing them makes you a better guesser, and explains why some verdicts about your own face sting.

We miss by years, not months

In one of the largest experiments on the topic, researchers asked people to estimate the age of nearly 4,000 passport photos. The average error was around eight years (Clifford, Watson and White, 2018). Passport photos are a hard test, with no context, no styling and a wide range of ages, but the lesson holds: a single guess is rough.

Young faces are easier

Age estimation is more accurate for young faces than for older ones, and younger people tend to be better at it than older people (Moyse and Brédart, 2014). Between 20 and 25 a face changes a lot; between 60 and 65 the changes are harder to see.

Everyone pulls guesses towards the middle

Young faces tend to be judged older and older faces younger, as if every guess were pulled towards the average age we're used to seeing. This bias is systematic: nearly everyone has it, so it doesn't disappear when you average many guesses.

The previous face changes your answer

The same passport study found that the biggest single source of error was the face seen just before. After an older face, the next guess drifts up; after a younger one, it drifts down. Together with the pull towards the middle, these two biases explained about 95% of the variation in the estimates.

We're best with our own generation

Both younger and older adults are more accurate and less biased when the face belongs to their own age group (Voelkle and colleagues, 2012). The same study found that expression matters: neutral faces were judged most accurately, and smiling faces were the most often judged younger than they were.

And AI?

Age estimation software learns from photos labelled by people, and it inherits our habits. A 2022 study found that the biases in human judgements of facial age were also present in current AI systems, and more exaggerated (Scientific Reports, 2022). More on that in AI age detector or real people.

Why a crowd does better

Many independent guesses fix part of the problem. Each voter comes to a face after a different one, so the pull of the previous face tends to cancel out. Own-age bias cancels too when voters of all ages take part. The pull towards the middle doesn't cancel, but it's the same for everyone, so comparisons between faces still work. That's why AgeDart uses the median of many votes, leaves out the ones far from the rest, and breaks your result down by the voter's age group: if people your age see you younger than twenty-year-olds do, you'll know.

We feel younger than we look

There's one more gap: between how old we look and how old we feel. In a study of 1,470 Danish adults, people over 40 felt about 20% younger than their real age (Rubin and Berntsen, 2006). So when strangers guess your age, the number can be fair and still feel too high. Curious anyway? Find out how old you look to real people, or sharpen your own eye with these tips to guess someone's age.