The human brain in numbersa linearly scaled-up primate brain

Suzana Herculano‐Houzel•View original

OverviewBalancedalloy voice

If you grew up on the usual brain lore, you probably heard three big claims. Humans have about one hundred billion neurons, roughly ten times as many glial cells, and a cortex that takes up more than eighty percent of the brain, which is why we’re uniquely smart. It’s a neat story. It’s also not how the numbers actually shake out when you start counting cells instead of repeating rules of thumb. Here’s the idea that changes the frame: brain size, by itself, is a lousy proxy for how many neurons a brain holds. Different mammalian lineages scale differently. So a big brain in one order doesn’t mean the same neuronal payload as a big brain in another. Suzana Herculano-Houzel pushed for a simple solution: stop inferring and start measuring. Build a cellular census across mammals and see where humans really land. To do that, she and Roberto Lent introduced a method called isotropic fractionation. It’s as straightforward as it sounds: you take a defined piece of brain tissue—say, the cerebral cortex or the cerebellum—break it down gently until you have a soup of free nuclei, stain the neuronal ones, and count. Because you’re not sampling little slivers and extrapolating, you can sum up absolute numbers for entire structures. Apply that across species—rodents, primates, insectivores—and you can compare cortex to cortex and cerebellum to cerebellum without leaning on brain size as a stand-in. What falls out of those counts is a clean split in the rules of growth. In primates, bigger brains scale almost linearly with neuron number. Double the neurons, and brain mass climbs a little over twofold. In rodents, brain mass outruns neuron number; as their brains get larger, they get sparser, and the ratio of glia to neurons rises. Mathematically, you can write it as a power law: in primates, brain mass grows as neuron number to about the one point zero six power; in rodents, it’s closer to one point five five. That gap is the quiet engine behind a lot of confusion about brain size and intelligence. If you push those rules, you get a striking practical consequence. Take a brain weighing around one and a half kilograms. A rodent built to that size would be predicted to hold something like twelve billion neurons. A primate at the same mass, by the primate rule, would be near ninety-three billion. That’s not a subtle difference. It’s the same ballpark as comparing a midsize city to a country. So where do humans land? On the primate line. As Azevedo and colleagues measured in 2009, the average adult male human brain, about one point five kilograms, contains eighty-six billion neurons and eighty-five billion non-neuronal cells. If you naïvely applied the primate scaling curves to that mass, you’d expect around ninety-three billion neurons and one hundred twelve billion non-neuronal cells. The actual human values come in roughly seven percent lower for neurons and about twenty-four percent lower for non-neuronal cells. That’s a rounding error compared to the rodent–primate gap. It says our brain isn’t a freak outlier; it’s a primate brain scaled up. The distribution inside that brain is just as telling. The human cerebral cortex is big in mass—about one thousand two hundred thirty-three grams, which is roughly eighty-two percent of the whole—but it houses a minority of the neurons: about sixteen billion, roughly nineteen percent of the brain’s total. The cerebellum, that dense, leafed structure tucked under the back of the cortex, is much lighter—about one hundred fifty-four grams—yet it contains about sixty-nine billion neurons. That pairing leads to a weird-to-your-intuition fact: the small cerebellum holds the lion’s share of neurons, while the massive cortex does most of its work with relatively fewer. Now let’s correct the glia story, because it’s been sticky. Glial cells—astrocytes, oligodendrocytes, microglia—do not outnumber neurons by ten to one in the human brain. Across the whole brain, glia constitute at most about half of all cells. In cortical gray matter, glia outnumber neurons, but only by less than two to one. In the cerebellum, neurons vastly dominate; there are at least twenty-five neurons for every glial cell. Those ratios shift with structure and lineage, but the blanket “ten to one” doesn’t survive a census. If you zoom back out across species, another pattern emerges that pokes at a common intuition: the fraction of a brain’s neurons that live in the cortex doesn’t track how big the cortex looks. Across fifteen of eighteen mammalian species sampled, the cortex contains between about thirteen and twenty-eight percent of all brain neurons, and that fraction stays uncorrelated with the cortex’s share of total brain size. Instead, cortex and cerebellum scale in concert. As the cortex gains neurons, the cerebellum does too, preserving a partnership we often underplay when we talk about cognition. The math behind those scaling differences is revealing without being onerous. In primates, the mass of the whole brain grows as neuron number to about one point zero six; the cortex alone grows as cortical neuron number to about one point zero eight; and the cerebellum keeps pace, near zero point ninety-nine. In rodents, the exponents are steeper—about one point five five for the whole brain and roughly one point seven four for the cortex—which encodes that “more tissue per neuron” trend as brains enlarge. Put more simply: primates pack neurons more tightly and sustain that packing as brains get bigger, while rodents spread out. Bring this back to the human case one more time, because it crystallizes the point. In a one point five kilogram brain, a rodent would hover near twelve billion neurons; a primate would sit around ninety-three billion. Humans are measured at eighty-six billion, just shy of that primate expectation. The brain’s mass alone can’t tell you which world you’re in; only a count can. What about those older metrics that tried to wrap cognition into a single number? The encephalization quotient, or EQ, was meant to do that by asking how big an animal’s brain is relative to its body. Humans famously top the charts, with EQ values usually cited between seven and eight across mammals and a bit over three if you confine the comparison to primates. The catch, as Herculano-Houzel has argued, is that EQ is sensitive to what species you put in the baseline and, within primates, human brains are only about ten percent above what you’d expect from a simple linear brain–body trend. In other words, EQ is a moving target. Neuron counts are not. Why prefer neurons? Because neurons are the computational units, and they don’t scale linearly in what they can do. Each neuron can connect to thousands of others. When you add neurons, you’re not just adding units, you’re adding potential pairings and pathways. That combinatorial explosion is invisible to brain size or EQ. It shows up directly in absolute neuron numbers and in how those numbers are distributed across interconnected systems like cortex and cerebellum. There’s a nice way to stress-test this with big-brained non-primates. Imagine a whale brain weighing around three and a half kilograms. If you forced rodent-like rules onto it, you’d predict on the order of twenty-one billion neurons. If it followed primate-like packing, you’d expect something like two hundred twelve billion. That hundred-billion swing hangs entirely on the scaling rule. Elephants tell the same story: huge brains don’t automatically mean more neurons. The lineage’s cellular economics decides. None of this is to say the method or the models are perfect. The comparative datasets sample many species but not all, and they sit on a phylogenetic tree—closely related species aren’t independent data points, a point others have rightly pushed on. The approach Herculano-Houzel took, comparing within orders, is meant to minimize that confound by asking how brain mass changes with neuron number inside a lineage rather than across all mammals at once. There’s also the question of variation. Across species, brains and neuron numbers span about five orders of magnitude. Within a single species, individuals vary much less—on the order of ten to fifty percent in size—so extrapolating fine-grained differences between two people from these broad scaling laws isn’t what the data are built for. One last calibration: this work isn’t claiming the cortex doesn’t matter, or that the cerebellum secretly runs the show. It’s a reminder that brains are networks of networks. The cortex is massive, metabolically expensive tissue doing high-dimensional transformations. The cerebellum is neuron-dense machinery calibrating timing, prediction, and coordination. Across primates, they scale together. When you grasp that, the human pattern—big cortex by mass, modest cortical share of neurons, cerebellum awash in neuronal units—looks less like an anomaly and more like a species-typical balance struck at large scale. Methodologically, that clarity hinged on counting. Isotropic fractionation is not glamorous, but it let teams like Azevedo’s in humans, Herculano-Houzel’s across primates and rodents, and Sarko and colleagues in other mammals tally entire structures without guesswork. Once you have the counts, the patterns about density, the ratios of glia to neurons, and scaling laws don’t need to be inferred from proxies. They’re in the numbers. So the payoff is this. Humans didn’t get a special brain recipe. We got a primate brain scaled up, landing us with the most neurons among primates because we also have the largest primate brain. Our cortex is huge by weight but holds about a fifth of the brain’s neurons. Our cerebellum, small by weight, holds most of them. Glia don’t swamp neurons by a factor of ten. And if you want to compare cognitive potential across species, count neurons—don’t let brain size or EQ do the talking for you. Where does this go next? The broad rules are in place; the frontier is finer grain. As Chen and others have suggested in related comparative work, we can now ask whether visual cortex scales with its own local rule, whether prefrontal areas buck the trend, and how subcortical hubs track their cortical partners. That’s the mosaic evolution question reframed at the cellular level. The promise is a map of brains not by how big they look from the outside, but by how many computational elements each part brings to the table, and how those parts scale together as minds get more complex.

It’s also not how the numbers actually shake out when you start counting cells instead of repeating rules of thumb.

Here’s the idea that changes the frame: brain size, by itself, is a lousy proxy for how many neurons a brain holds. Different mammalian lineages scale differently. So a big brain in one order doesn’t mean the same neuronal payload as a big brain in another.

Suzana Herculano-Houzel pushed for a simple solution: stop inferring and start measuring. Build a cellular census across mammals and see where humans really land.

To do that, she and Roberto Lent introduced a method called isotropic fractionation. It’s as straightforward as it sounds: you take a defined piece of brain tissue—say, the cerebral cortex or the cerebellum—break it down gently until you have a soup of free nuclei, stain the neuronal ones, and count. Because you’re not sampling little slivers and extrapolating, you can sum up absolute numbers for entire structures.

Apply that across species—rodents, primates, insectivores—and you can compare cortex to cortex and cerebellum to cerebellum without leaning on brain size as a stand-in.

What falls out of those counts is a clean split in the rules of growth. In primates, bigger brains scale almost linearly with neuron number. Double the neurons, and brain mass climbs a little over twofold.

In rodents, brain mass outruns neuron number; as their brains get larger, they get sparser, and the ratio of glia to neurons rises. Mathematically, you can write it as a power law: in primates, brain mass grows as neuron number to about the one point zero six power; in rodents, it’s closer to one point five five. That gap is the quiet engine behind a lot of confusion about brain size and intelligence.

If you push those rules, you get a striking practical consequence. Take a brain weighing around one and a half kilograms. A rodent built to that size would be predicted to hold something like twelve billion neurons.

A primate at the same mass, by the primate rule, would be near ninety-three billion. That’s not a subtle difference. It’s the same ballpark as comparing a midsize city to a country.

So where do humans land? On the primate line. As Azevedo and colleagues measured in 2009, the average adult male human brain, about one point five kilograms, contains eighty-six billion neurons and eighty-five billion non-neuronal cells.

If you naïvely applied the primate scaling curves to that mass, you’d expect around ninety-three billion neurons and one hundred twelve billion non-neuronal cells. The actual human values come in roughly seven percent lower for neurons and about twenty-four percent lower for non-neuronal cells. That’s a rounding error compared to the rodent–primate gap. It says our brain isn’t a freak outlier; it’s a primate brain scaled up.

The distribution inside that brain is just as telling. The human cerebral cortex is big in mass—about one thousand two hundred thirty-three grams, which is roughly eighty-two percent of the whole—but it houses a minority of the neurons: about sixteen billion, roughly nineteen percent of the brain’s total. The cerebellum, that dense, leafed structure tucked under the back of the cortex, is much lighter—about one hundred fifty-four grams—yet it contains about sixty-nine billion neurons.

That pairing leads to a weird-to-your-intuition fact: the small cerebellum holds the lion’s share of neurons, while the massive cortex does most of its work with relatively fewer.

Now let’s correct the glia story, because it’s been sticky. Glial cells—astrocytes, oligodendrocytes, microglia—do not outnumber neurons by ten to one in the human brain. Across the whole brain, glia constitute at most about half of all cells.

In cortical gray matter, glia outnumber neurons, but only by less than two to one. In the cerebellum, neurons vastly dominate; there are at least twenty-five neurons for every glial cell. Those ratios shift with structure and lineage, but the blanket “ten to one” doesn’t survive a census.

If you zoom back out across species, another pattern emerges that pokes at a common intuition: the fraction of a brain’s neurons that live in the cortex doesn’t track how big the cortex looks. Across fifteen of eighteen mammalian species sampled, the cortex contains between about thirteen and twenty-eight percent of all brain neurons, and that fraction stays uncorrelated with the cortex’s share of total brain size. Instead, cortex and cerebellum scale in concert.

As the cortex gains neurons, the cerebellum does too, preserving a partnership we often underplay when we talk about cognition.

The math behind those scaling differences is revealing without being onerous. In primates, the mass of the whole brain grows as neuron number to about one point zero six; the cortex alone grows as cortical neuron number to about one point zero eight; and the cerebellum keeps pace, near zero point ninety-nine. In rodents, the exponents are steeper—about one point five five for the whole brain and roughly one point seven four for the cortex—which encodes that “more tissue per neuron” trend as brains enlarge.

Put more simply: primates pack neurons more tightly and sustain that packing as brains get bigger, while rodents spread out.

Bring this back to the human case one more time, because it crystallizes the point. In a one point five kilogram brain, a rodent would hover near twelve billion neurons; a primate would sit around ninety-three billion. Humans are measured at eighty-six billion, just shy of that primate expectation. The brain’s mass alone can’t tell you which world you’re in; only a count can.

What about those older metrics that tried to wrap cognition into a single number? The encephalization quotient, or EQ, was meant to do that by asking how big an animal’s brain is relative to its body. Humans famously top the charts, with EQ values usually cited between seven and eight across mammals and a bit over three if you confine the comparison to primates.

The catch, as Herculano-Houzel has argued, is that EQ is sensitive to what species you put in the baseline and, within primates, human brains are only about ten percent above what you’d expect from a simple linear brain–body trend. In other words, EQ is a moving target. Neuron counts are not.

Why prefer neurons? Because neurons are the computational units, and they don’t scale linearly in what they can do. Each neuron can connect to thousands of others.

When you add neurons, you’re not just adding units, you’re adding potential pairings and pathways. That combinatorial explosion is invisible to brain size or EQ. It shows up directly in absolute neuron numbers and in how those numbers are distributed across interconnected systems like cortex and cerebellum.

There’s a nice way to stress-test this with big-brained non-primates. Imagine a whale brain weighing around three and a half kilograms. If you forced rodent-like rules onto it, you’d predict on the order of twenty-one billion neurons.

If it followed primate-like packing, you’d expect something like two hundred twelve billion. That hundred-billion swing hangs entirely on the scaling rule. Elephants tell the same story: huge brains don’t automatically mean more neurons. The lineage’s cellular economics decides.

None of this is to say the method or the models are perfect. The comparative datasets sample many species but not all, and they sit on a phylogenetic tree—closely related species aren’t independent data points, a point others have rightly pushed on. The approach Herculano-Houzel took, comparing within orders, is meant to minimize that confound by asking how brain mass changes with neuron number inside a lineage rather than across all mammals at once.

There’s also the question of variation. Across species, brains and neuron numbers span about five orders of magnitude. Within a single species, individuals vary much less—on the order of ten to fifty percent in size—so extrapolating fine-grained differences between two people from these broad scaling laws isn’t what the data are built for.

One last calibration: this work isn’t claiming the cortex doesn’t matter, or that the cerebellum secretly runs the show. It’s a reminder that brains are networks of networks. The cortex is massive, metabolically expensive tissue doing high-dimensional transformations.

The cerebellum is neuron-dense machinery calibrating timing, prediction, and coordination. Across primates, they scale together. When you grasp that, the human pattern—big cortex by mass, modest cortical share of neurons, cerebellum awash in neuronal units—looks less like an anomaly and more like a species-typical balance struck at large scale.

Methodologically, that clarity hinged on counting. Isotropic fractionation is not glamorous, but it let teams like Azevedo’s in humans, Herculano-Houzel’s across primates and rodents, and Sarko and colleagues in other mammals tally entire structures without guesswork. Once you have the counts, the patterns about density, the ratios of glia to neurons, and scaling laws don’t need to be inferred from proxies. They’re in the numbers.

So the payoff is this. Humans didn’t get a special brain recipe. We got a primate brain scaled up, landing us with the most neurons among primates because we also have the largest primate brain.

Our cortex is huge by weight but holds about a fifth of the brain’s neurons. Our cerebellum, small by weight, holds most of them. Glia don’t swamp neurons by a factor of ten.

And if you want to compare cognitive potential across species, count neurons—don’t let brain size or EQ do the talking for you.

Where does this go next? The broad rules are in place; the frontier is finer grain. As Chen and others have suggested in related comparative work, we can now ask whether visual cortex scales with its own local rule, whether prefrontal areas buck the trend, and how subcortical hubs track their cortical partners.

That’s the mosaic evolution question reframed at the cellular level. The promise is a map of brains not by how big they look from the outside, but by how many computational elements each part brings to the table, and how those parts scale together as minds get more complex.

Related lectures