The famous limit of 150 friends a human brain can supposedly hold turns out to rest on a calculation whose own confidence interval runs from 2 to 520 — too wide to support any specific number, let alone the one everyone quotes

The number 150 shows up in strange places for something that started as a footnote about primate brains. A Facebook executive once called it “one of the magic numbers in group sizes,” telling Quartz that “weird stuff starts to happen” once an organization crosses it and needs real structure to hold itself together. It gets cited in team-building workshops, HR decks, and roughly every “how to build community online” post written in the last decade. All of it traces back to one anthropologist’s estimate of how many stable relationships a human brain can actually track.

What almost none of those citations mention is that the number’s own statistical foundation, when a separate team of researchers went back and rebuilt the original calculation properly, turned out to be barely a number at all.

Where 150 actually came from

The anthropologist Robin Dunbar arrived at 150 in the 1990s by comparing neocortex size across primate species and looking for a relationship between brain structure and typical group size. Humans, plotted along that same regression line, landed around 150. The idea was elegant enough to survive three decades of pop-science retellings largely intact: bigger brain region, bigger stable social circle, and a hard ceiling somewhere around a number most people could remember.

The trouble, according to a 2021 reanalysis, was never really the concept. It was the arithmetic underneath it.

What happens when you rerun the numbers properly

The evolutionary biologist Patrik Lindenfors, at Stockholm University’s Centre for Cultural Evolution, working with Andreas Wartel and Johan Lind, went back to the primate brain-and-group-size data Dunbar’s original estimate rested on and reran it using statistical methods that account for uncertainty properly — Bayesian models and phylogenetic regressions that don’t just spit out a single tidy number, but a range of numbers the data can actually support. Their results, published in Biology Letters, were not subtle. Depending on which model they used, the 95 percent confidence interval for a human group-size estimate came out somewhere between 4 and 520 people in one analysis, and somewhere between 2 and 336 in the other — a spread so wide that coverage at the time simply described it as running from 2 to 520 people.

For context on how enormous that actually is: a confidence interval that wide fails to rule out a “stable social group” of two people, or one of five hundred. A number that can’t distinguish between a marriage and a small university lecture hall isn’t imprecise. It’s simply not telling you anything.

The reanalysis wasn’t a smaller or sloppier version of the original either — if anything it worked with more data than Dunbar had access to in the 1990s, drawing on measurements from 71 primate species with complete brain and neocortex data, run through their best-fitting model twice over. The single point estimate that model spat out, when forced to produce one number rather than a range, landed around 69 — less than half of 150. But the team was explicit that reporting that number on its own would have repeated the same mistake they were trying to correct: a point estimate without its uncertainty is exactly how 150 escaped its original methods section and turned into a rule people build companies around.

A range running from 2 to 520 isn’t a more careful version of 150. It’s the data admitting no single number was ever really in there.

What the researchers actually said about their own finding

What I found most disarming about this wasn’t the statistics, it was how plainly the researchers described the gap between what the data shows and what people want to hear. Andreas Wartel, quoted alongside the study, put the core finding as bluntly as a scientist ever does: “It is not possible to make an estimate for humans with any precision using available methods and data.” Lindenfors went further, naming exactly why the flawed number survived so much scrutiny for so long: “I think Dunbar’s number is widely spread, also among researchers, since it’s so easy to understand. Our claim that it is not possible to calculate a number is not quite as entertaining.”

That line is doing more work than it looks like it’s doing. Read past the self-deprecation and it’s a fairly precise diagnosis of why bad statistics survive in public life at all: a specific, memorable number will always out-compete an honest, unglamorous range, regardless of which one the underlying data actually supports. “Somewhere between 2 and 520, we genuinely can’t say” doesn’t fit on a slide. “150” does.

Why a number this uncertain still runs entire org charts

I’ve cited Dunbar’s number myself, in a lecture, years ago, as a tidy way to illustrate that human social capacity has limits — and nobody in that room, including me, thought to ask what was actually holding the number up. That’s the uncomfortable part of revisiting it now, as someone who works in psychology research: it wasn’t a fringe factoid I picked up carelessly. It was sitting in textbooks, cited in papers, treated as settled enough that checking its confidence interval never occurred to me as a necessary step.

I don’t think the people restructuring their companies around Dunbar’s number are being careless, exactly. A round, specific figure is genuinely useful as a design constraint, even a slightly made-up one — it gives a team a target to plan around instead of an open-ended “somewhere in between.” Plenty of good decisions get made using rules of thumb that were never rigorously derived.

What’s worth questioning is the confidence that gets borrowed along with this particular number — the sense that because it’s attributed to neuroscience and comes with a clean regression line behind it, it must outrank an ordinary guess.

According to the researchers who actually rebuilt the calculation, the data underneath it can’t even tell 2 from 520, which makes it considerably less reliable than most guesses.

What actually survives the critique

Nevertheless, this doesn’t mean that Dunbar’s broader observation was worthless. The idea that human relationships come in layers — a small circle of real intimacy, a wider ring of people you’d call friends, an outer ring of people you’d recognize and greet — holds up fine without a specific number attached to any of the rings. What doesn’t survive is the specific claim that neuroscience handed us a hard ceiling at 150, precise enough to plan a company or a wedding guest list around.

What I keep sitting with is less about friendship and more about how many other “science says” numbers I’ve absorbed without asking where the confidence interval went.

Most of them, I’d guess, never had one attached in the first place. Somewhere between finding that out and forgetting it again, there’s probably a number worth tracking too.