Dunbar's
Number
The famous "150" is a hypothesis about the scale of an active, stable social network - not a hard cap on acquaintances, followers, organizations, or human cooperation.
Relationships consume unequal time.
Allocate weekly social-maintenance time and change the average intensity of the network. This deliberately simple budget model illustrates a mechanism; it does not diagnose your personal capacity or derive Dunbar's number from first principles.
This setting reproduces the familiar central estimate, but the underlying evidence supports variation and layered intensity rather than a universal cutoff.
150 is a central estimate for one kind of network.
Dunbar's number is commonly defined as the approximate number of people with whom one person can maintain stable social relationships - relationships in which identities, histories, and mutual positions remain meaningful. It does not count every face recognized, every address-book entry, every customer, or every follower.
Dormant schoolmates and one-time contacts may remain known without occupying the same maintenance layer.
The tie persists long enough to support mutual knowledge, obligation, or predictable interaction.
Audience reach and celebrity recognition are different network objects from reciprocal personal ties.
The useful idea is not that friendship stops at person 151. It is that social capacity has costs, and close relationships occupy more of it than weak ones.
An extrapolation from primate comparative data.
Dunbar's 1992 paper examined relationships between relative neocortex size and typical social group size across nonhuman primates. A fitted relationship was then extrapolated to humans, producing a prediction near 150. The argument formed part of the broader social-brain hypothesis: cognitively demanding social life may be associated with the evolution of larger brains.[1]
Every step creates uncertainty. Species are not statistically independent in the ordinary sense because they share evolutionary history. Brain regions and group sizes can be measured in multiple ways. Group size varies within species and with ecology. Most importantly, predicting beyond the sampled primate range is an extrapolation, not a direct human measurement.
The 2021 reanalysis by Lindenfors, Wartel, and Lind obtained strongly method-dependent estimates and very broad uncertainty, concluding that one cognitive limit cannot be derived reliably by this comparative route.[5]
Social networks look more like nested circles than one ceiling.
Later work associated with Dunbar describes a series of approximately threefold layers. The innermost group receives the most time and emotional investment; progressively larger layers receive less. Values such as 5, 15, 50, and 150 are population-level approximations, not required personal totals.[3]
Layer membership changes. Life transitions, distance, conflict, partnership, work, illness, and migration can move people inward or outward. Network turnover is part of the phenomenon.
Brain capacity is only one possible bottleneck.
Relationships involve identity, history, alliances, intentions, obligations, and third-party ties.
Conversation, shared activity, care, and repair require repeated attention.
Trust and attachment depend on investment that cannot be replicated by increasing contact lists.
Groups face scheduling, common knowledge, conflict, and communication overhead as they grow.
Mobility, settlement, subsistence, risk, institutions, and communication technologies reshape feasible networks.
Kinship, friendship, teams, associations, and community boundaries differ across societies.
The broad constraint is plausible; the unique number is disputed.
Recurring scales and time allocation
- Some personal-network datasets show layered distributions of contact and emotional closeness.
- Historical and ethnographic group sizes are sometimes reported near predicted layers.
- A large Twitter conversation study found an active communication range around 100-200 despite much larger follower networks.[4]
Methods and definitions move the result
- Comparative regressions depend on brain measure, species sample, group-size statistic, phylogenetic method, and regression choice.
- Human communities are culturally organized and cannot be inferred from primate anatomy alone.
- Archaeological, historical, online, and organizational groups may measure co-residence, recognition, interaction, or administration - not the same tie.
- The 2021 reanalysis found estimates too unstable and intervals too broad to justify a single derived limit.[5]
Platforms expand reach more easily than relationship depth.
Often one-way, low-maintenance, and algorithmically mediated. It should not be compared directly with an active personal network.
Knowing how to reach someone is not equivalent to maintaining a stable reciprocal tie.
Repeated replies and direct exchanges better approximate active contact, though context and meaning remain hidden.
Who provides help, trust, or care is closer to the strong-tie claim, but harder to observe at platform scale.
Digital tools reduce search, scheduling, and transmission costs. They may preserve weak ties across distance and make dormant ties easier to reactivate. They do not remove finite time or guarantee reciprocity. A million followers can coexist with a small support clique.
Do not turn 150 into an org-chart commandment.
Direct collaboration
Most tightly interdependent teams should be far smaller than 150 because coordination intensity is high.
DESIGN FOR: clear tasks + dense trustShared identity
A larger unit can maintain recognition and common purpose without every pair collaborating frequently.
DESIGN FOR: subgroups + shared contextInstitutional scale
Organizations exceed personal-network scales through roles, processes, records, markets, law, hierarchy, and modular structure.
DESIGN FOR: interfaces + governanceVoluntary association
Events, rituals, norms, places, and communication channels can coordinate many weak and overlapping ties.
DESIGN FOR: layered participationWho must know whom? Which ties require trust? Where can structure replace personal familiarity?
NOT AS A RULENever split a healthy organization solely because a memorable number was reached.
What the number does not license.
"Humans can know only 150 people."
Recognition and acquaintance layers can be much larger; the claim concerns stable active ties.
"Person 151 must displace a friend."
Networks are graded, dynamic, overlapping, and individually variable.
"All groups should contain 150."
Task, coordination intensity, hierarchy, subgrouping, ecology, and institution change viable size.
"Online data prove the brain limit."
Platform rules, sampling, algorithms, bots, and observable behavior complicate inference.
"A primate regression settles human culture."
Comparative biology is relevant evidence, not a complete account of human social organization.
"150 is precise."
Study design and statistical method produce large uncertainty; report ranges and definitions.
Human ties inside larger systems.
Metcalfe's Law
Possible digital links grow far faster than the time available for stable reciprocal relationships.
Parkinson's Law
Unit size alone cannot explain workload expansion, hierarchy, or coordination drag.
Strong and weak ties
Different ties provide support, information, identity, opportunity, and bridging across communities.
Brooks's Law
Interdependent work gains communication paths as people are added, making team structure important.
Sources and further reading.
The original hypothesis, later refinements, online evidence, and direct statistical criticism are all included.
- R. I. M. Dunbar (1992) - Neocortex Size as a Constraint on Group Size in PrimatesThe original comparative analysis from which the human group-size estimate emerged.Journal of Human Evolution, 22(6), 469-493
- R. I. M. Dunbar (1993) - Coevolution of Neocortical Size, Group Size and Language in HumansDevelops the social-grooming and language argument for maintaining unusually large human groups.Behavioral and Brain Sciences, 16(4)
- R. I. M. Dunbar (2018) - The Anatomy of FriendshipA review of layered social networks, relationship quality, time allocation, and individual variation.Trends in Cognitive Sciences, 22(1), 32-51
- Goncalves, Perra & Vespignani (2011) - Modeling Users' Activity on Twitter NetworksA large conversation-network study reporting an active communication range around 100-200.PLOS ONE, 6(8), e22656
- Lindenfors, Wartel & Lind (2021) - 'Dunbar's Number' DeconstructedA phylogenetically informed reanalysis finding method-sensitive estimates and very broad uncertainty.Biology Letters, 17, 20210158
- de Ruiter, Weston & Lyon (2011) - Dunbar's Number: Group Size and Brain Physiology in Humans ReexaminedA critical anthropological examination of the proposed relationship and its application to human groups.American Anthropologist, 113(4), 557-568