Dunbar’s Number: Can We Really Have Only 150 Friends?

Open your phone’s contact list and count the names. Now imagine a different list: the people you would call after a terrible day. Then a third: everyone you would be pleased to catch up with at a party.

You could produce three very different numbers without gaining or losing a single relationship. You changed what you were counting.

That is the first thing to understand about Dunbar’s number, the famous claim that people can keep up about 150 relationships. The number has some support as a rough description of certain extended personal networks. But the evidence does not establish a universal limit of 150 friends, and it does not tell us that a company should stop growing at 150 employees.

To see why, we need to separate three questions: where the number came from, what later studies actually measured, and what would justify turning an average into a rule.

What is Dunbar’s number?

Dunbar’s number is the proposal that humans can maintain about 150 stable social relationships at a time. It is named after Robin Dunbar, the British anthropologist and evolutionary psychologist at the University of Oxford who proposed it. His idea is that the mental work of maintaining relationships limits the size of our social world. That work includes remembering people, understanding how they relate to each other, and keeping track of life in a group.

In this theory, 150 refers to an extended network of stable relationships. It does not mean 150 best friends. Family members and more distant social ties can be part of that network. Nor does it include everyone whose name you recognize.

The idea deserves a hearing. Remembering a name is one thing. Keeping up a relationship with that person is another, and it takes time and attention. The harder question is whether our brains set a specific numerical limit, and whether 150 is that limit. In a 1992 paper, Dunbar argued that in primates, the size of the neocortex limits how many relationships an animal can keep track of, and so limits how large a group can stay together. A year later, he applied that argument to humans.

Where did the number 150 come from?

The original calculation began with other primates.

Dunbar compared average group size with the relative size of the neocortex, part of the brain’s outer layer. His measure was the neocortex ratio: the volume of the neocortex divided by the volume of the rest of the brain. He used averages for 36 primate genera (a genus is a group of closely related species).

The relationship was strong. Primates with relatively larger neocortices tended to live in larger groups, and the ratio accounted for about 76% of the variation in group size across those genera. Dunbar fitted an equation to that relationship, then plugged in the human ratio of 4.1. The result was 147.8, which became the familiar 150. In the 1993 paper that made the human prediction, he also compared that figure with real human groups, including mid-sized groupings in traditional societies, Hutterite farming communities and basic army units.

But the human ratio lay outside the range used to build the equation. It was about 50% larger than the ratio for any other primate. Predicting from there required an extrapolation: assuming that a pattern observed within one range continues outside it. Dunbar acknowledged this in the paper and called the step exploratory. His own 95% confidence interval for the human prediction ran from about 100 to 231.

Imagine measuring the height and weight of a group of children, finding a relationship between the two, and using it to predict the weight of a much taller adult. You can do the calculation. Whether the relationship still holds for adults is a separate question. The same issue arises when an equation fitted to other primates is extended to humans.

There is a second step to justify. Dunbar’s equation used average group sizes. Even if it predicted the average human network perfectly, that would not automatically reveal the largest network an individual could maintain.

Suppose the average household on a street contained three people. You would not conclude that a fourth person could not move in. An average describes what you observed. A capacity limit makes a further claim about what is possible.

Dunbar had a reason for using averages. Primate groups tend to split once they grow too large, so he treated a species’ average group size as a better guide to its limit than the largest group ever seen. That is an argument about groups, though. It does not tell you how many relationships any one person can handle.

What did studies of human relationships find?

The primate equation is only part of the evidence. Researchers have also counted people’s actual social networks, and some results come out close to 150. These studies deserve attention. So do their counting rules.

The Christmas-card study

In a 2003 study, Russell Hill and Robin Dunbar used Christmas cards to map social networks. Their reasoning: in Western societies, the card season is the one time of year many people make an effort to contact everyone they value keeping in touch with. They collected 43 questionnaires, one per household, all completed by white British respondents.

A card could go to an individual, a couple, or a whole family. The researchers counted everyone in each recipient household. They also added the members of the sender’s own household and the people respondents usually sent cards to but expected to see over Christmas instead. With this broad count, the average network contained 153.5 people.

That looks remarkably close to the original prediction. But the counts varied a lot between households: the standard deviation was 84.5 people, more than half the average. A result near 150 described the average, not a size everyone shared.

The researchers also made a narrower count: only the people respondents said they actively contacted, leaving out the rest of each recipient’s household. Only 22 of the 43 questionnaires recorded that distinction. For those 22, the average was 124.9.

Consider a card sent to an old colleague, her husband, and their two children. Counting everyone in the household adds four people. Counting only the colleague you actually keep in touch with adds one. Both counts are legitimate. They answer different questions. And because the two averages come from different sets of questionnaires (43 versus 22), the counting rule is not the only reason they differ.

The broad result is evidence that extended social networks can average around 150 in a particular setting. It cannot tell us that every person has room for exactly that many relationships.

The phone-record study

A much larger study looked at a full year of mobile-phone calls. Pádraig Mac Carron, Kimmo Kaski and Robin Dunbar analyzed 2007 call records from a European phone company and published the results in 2016 as Calling Dunbar’s Numbers. To screen out business and casual calls, they counted a contact only if calls went in both directions.

For their main analysis, they kept only the 26,680 users who had at least 100 such contacts, out of about 6 million customers whose calls they could fully see. Within that group, the average was 129.9 contacts.

The selection rule matters. An average for people who already have at least 100 reciprocal phone contacts is not an average for everyone who owns a phone. The researchers had a sensible reason for the cutoff: people who use their phones only to arrange meetings or handle emergencies have call records that show only part of their networks. But the cutoff also means the 129.9 describes people with at least 100 two-way phone contacts, not people in general.

Phone calls also capture only part of a social life, and the authors said so. A friend you see every day might rarely appear in your call history. Someone you call regularly might be a business contact. Requiring calls in both directions helps narrow the measure, but it does not turn every connection into a lasting friendship.

Together, these studies help explain why 150 remains a useful reference. They also show why it would be a mistake to swap in a supposedly more accurate number, such as 125 or 130. Those figures describe different samples and different kinds of connection. The popular claim that a habit takes 66 days to form has a similar history: the figure is a modeled median from a selected group of study participants.

How many people do you know? That is a different question.

Count more loosely and you get a much larger number. In US survey research published in 2001, Christopher McCarty and colleagues estimated an average personal network of about 291 people. They used a nationally representative US sample. Their two methods asked people how many they knew in groups of known size, such as diabetics or Native Americans, and in relationship categories, such as family or coworkers.

Their definition of knowing someone was broad. You know them and they know you by sight or by name, you could contact them, they live in the US, and you have had some contact in the past two years. That can include acquaintances who would never make a list of relationships you actively keep up.

Your former neighbor might qualify under that definition even if you have spoken only once since moving away. The same person might not qualify when the question is whom you currently rely on for support.

So a count above 150 does not, by itself, disprove a limit defined for a narrower kind of relationship. And an average near 150 does not prove that limit. Before comparing the answers, check that the studies asked the same question.

Does social media change Dunbar’s number?

Social media makes the counting problem easy to see. A listed friend, a follower, and someone you regularly exchange messages with are three different measures.

In two UK surveys Dunbar published in 2016, people reported an average of about 155 Facebook friends in one sample and 183 in the other. The first sample was 2,000 adults who used social media regularly. The second was 1,375 professionals with full-time weekday jobs, not chosen for their social-media use. People picked their friend count from a list of ranges rather than giving an exact number. On average, people in the first sample said only about 28% of their Facebook friends were genuine, close friends.

Dunbar read these results as evidence of a cognitive limit that online media cannot get around. But a self-reported friend count is not a direct test of how many lasting relationships a person can keep up.

Bruno Gonçalves, Nicola Perra and Alessandro Vespignani looked at interaction instead of friend lists. In a 2011 study, they built a network from six months of Twitter conversations involving 1.7 million users, linking each user to the people they replied to. As users replied to more people, the average number of replies each contact received rose at first, then peaked when a user was replying to somewhere between 100 and 200 people. Past that point, attention was spread thinner. The authors interpreted this as support for Dunbar’s number.

The observation is useful: people did not spread their attention evenly across an ever-growing set of contacts. But replies on one platform do not reveal a person’s whole offline network, and they do not establish a sharp limit on friendship. The researchers also built a computer model that reproduced the pattern. That model assumed limited attention from the start: each simulated user could handle only so many messages at once.

Where the peak landed depended on how many messages the model let each user handle. The model shows that limited attention can produce the pattern. It does not independently prove a biological ceiling.

Technology can make a message easier to send. That does not make a follower count a measure of how many relationships someone maintains.

What about the circles of 5, 15, 50, and 150?

Dunbar’s theory also describes social relationships in layers, often summarized as circles of roughly 5, 15, 50, and 150 people. The inner circles contain closer relationships; the outer circles add more distant ones. Each circle is about three times the size of the one inside it.

These are cumulative counts. The five people in the smallest circle are included in the circle of 15. You do not add all four figures together.

The phone-record study found evidence of layers, but the exact results depended on which users were included and how the researchers sorted their calls. The researchers tried several methods for grouping each user’s contacts by how often they called them. With one method, the most common result was four groups, with cumulative averages of about 4, 11, 30, and 129 contacts. But that method found four groups for only about 27% of the selected users. Another 22% fell into five groups, and the other methods produced different arrangements.

In the published paper, the authors concluded that the innermost and outermost layers matched earlier studies well, while the layers in between varied a lot. When the researchers forced every user into four groups, each layer was on average about 3.3 times the size of the one inside it.

The useful finding is that people spread their attention unevenly across relationships. Those results do not show that everyone’s social life naturally divides into the same numbered circles.

Has Dunbar’s number been debunked?

The answer depends on which claim you mean.

In 2021, Patrik Lindenfors, Andreas Wartel and Johan Lind reran the primate-brain calculation with larger, more recent primate datasets and methods that account for how closely related the species are. Their conclusion was blunt: this method cannot produce a reliable number. Depending on the statistical method and the brain measure, their estimates of average human group size ran from about 16 to 109. Across their models, the 95% intervals stretched from as low as 2 to as high as 520. Their preferred model put the average at about 69, with an interval of about 4 to 292.

Then they went further. Drawing on other lines of research as well, they argued that there is no hard cognitive limit on human sociality. They said they hoped the number would drop out of use in science and the popular media.

Dunbar and Susanne Shultz answered in a 2023 review of comparative brain research. They argued that Lindenfors and colleagues had fitted a single line through all primates, even though primates fall into separate grades: clusters of species that sit on different lines. If each grade rises steeply but the grades sit at different heights, one line drawn through all of them comes out flatter than any of them, and so predicts too low a number for humans. Fitted only to the grade that contains the apes, Dunbar and Shultz said, the equation predicts about 152, close to the roughly 154 they report as the observed human average. (A line through the apes alone gives about 140.)

They also argued that the wide intervals Lindenfors reported describe where a single case could fall, not how precisely the average can be predicted. Even on their preferred line, the range for the average ran from about 71 to 195.

The numerical dispute turns on statistical choices: which primates go into the equation, which line you fit, and which kind of uncertainty you report.

Keep two claims apart. The first is that people’s extended networks often average somewhere near 150. Some human studies support that, and it can survive criticism of the original primate equation. The second is that everyone has the same capacity, so that friend number 151 must push someone else out. The evidence does not support that. Even in the Christmas-card study, respondents’ networks varied widely around the 153.5 average, with a standard deviation of 84.5.

Nor does uncertainty about the number mean social capacity is unlimited. Dropping a precise limit does not give anyone unlimited time or attention. And the evidence reviewed here does not support a single replacement number.

Should a business stop growing at 150 people?

That would require a different kind of evidence.

A personal network is organized around one person. A company consists of people with different jobs, teams, and working relationships. Its employees also have relationships outside work. Moving from “a person’s extended network may contain around 150 people” to “an organization works best with 150 employees” changes both what is being counted and what success means.

Imagine two organizations with the same headcount. In one, nearly every decision requires agreement across the whole staff. In the other, teams can complete most work independently. The coordination demands differ even though the number of employees is identical. A useful staffing rule would need to account for that difference.

A 2020 paper by Bruce West, Paolo Grigolini, Dunbar and four colleagues ran two abstract computer models of networks. In one, simulated individuals switch between two choices while copying their neighbors. The other is a model of flocking. In both, a measure of the network’s complexity peaked at around 150 members.

Those were simulations. They did not test whether real companies perform better at that size.

At least one large organization has acted on the number anyway. In 2007, the Swedish Tax Agency (Skatteverket) announced its biggest reorganization in years, moving work from small towns to larger towns and cities. Economies of scale were a large part of the reason. But according to an internal report seen by the Swedish news agency TT, the agency also set an upper limit of 150 employees per office. The report justified the limit with the primate studies and quoted the same 147.8 that Dunbar’s equation produced.

In 2011, Jan de Ruiter, Gavin Weston and Stephen Lyon, writing in American Anthropologist, argued that whether the reorganization succeeded could, in principle, be tested. A search of English and Swedish sources turned up no published evaluation that did. Adopting a size rule is a decision, not evidence that the rule works.

Before recommending a hiring limit, we would need evidence about actual organizations: their work, how people coordinate, what happens when they grow or split, and the benefits and costs of those changes. The studies discussed here do not establish a best company size or a measured benefit from enforcing 150. The evidence on nudges shows the same problem: the hard step is the jump from a striking result to a dependable design recipe.

What should you do with the number?

Use it to ask a better question about your social life, not to set a target.

If you feel stretched, consider which relationships you are trying to maintain and what maintaining them involves. A weekly conversation with a close friend, an occasional family visit, and an annual exchange with an old colleague make different demands. Counting each of them once does not capture those demands.

For a team or community, start with the work people need to do together. Are decisions slow because too many people must agree? Are responsibilities unclear? Do people lack time to maintain useful working relationships? These questions concern the demands of the work. A headcount alone cannot answer them.

Return to those three lists: your phone contacts, the people you would call after a terrible day, and the people you would enjoy seeing at a party. A number is informative only when you know which list it describes. There is no reason to make all three come to 150.