I wrote this in November of 2015, when I had just finished pulling the whole USDA nutrient table into Mathematica and dividing one column by another. The original post was a paragraph of exclamations and a link to a spreadsheet. The spreadsheet is still here — but the link had rotted, so I have rebuilt the page around the data instead of merely pointing at it. Every one of the 8,789 foods is now searchable below, and the file is downloadable in three formats.
The numbers themselves are a snapshot of the USDA National Nutrient Database for Standard Reference as it stood in late 2015. Standard Reference has since been retired; its final edition, SR Legacy, was folded into FoodData Central in 2019, which is where you should go for current figures. I have left the 2015 values untouched, because a decade-old snapshot is a perfectly good thing to be, as long as it says so out loud.
I did this as a quick project for a friend who wanted to know where to put her money and her appetite to get the most protein. The USDA publishes protein and energy for thousands of foods; dividing the one column by the other takes about four seconds. Reading the result took considerably longer, and I have been coming back to it for a decade.
How to read the number
Each food carries a single figure: grams of protein per 100 kilocalories. Soy protein isolate, at the top of the list, gives 27.5 grams of protein for every 100 calories it costs you. Cane sugar, at the very bottom, gives zero. Raw spinach gives 12.4. A slice of commercially made blueberry pie gives 0.78.
It is worth being clear about what this ratio does and does not see. It is not protein per gram, and it is not protein per serving. It measures protein relative to energy, which means it quietly rewards two things: leanness and water. Pacific cod scores 24.4 not because it is unusually protein-dense by weight — it is about 18% protein — but because the rest of it is mostly water and almost none of it is fat. Almonds are roughly 21% protein by weight, nearly the same, yet they land at rank 4,751 with 3.65, because they arrive with a great deal of oil. A ratio like this one has an opinion, and the opinion is: lean, wet, and unsweetened.
That is a useful opinion if you are trying to eat more protein without eating more calories. It is a misleading one if you are trying to eat enough of anything at all, which is why I would not hand this list to someone underweight, or elderly, or recovering from illness, without a good deal of conversation first.
The top of the list is a joke, and the joke is division
The first thirty-nine entries are not foods that beat soy protein isolate. They are entries where the denominator went to zero and Mathematica said so, politely, in its own vocabulary. Thirty-six of them came back Indeterminate — no protein, no calories, 0/0. That is table salt, baking soda, tap water, Perrier, Evian, brewed hibiscus tea, and a small parade of diet sodas. Three came back ComplexInfinity — a trace of protein divided by zero calories, which is to say decaffeinated brewed coffee and two low-calorie sodas.
I have kept all thirty-nine, since deleting inconvenient rows is how data goes bad, they now sit in their own drawer beneath the table rather than crowning the ranking, and every rank quoted on this page counts only the 8,750 foods where the division actually produced a number. The related and subtler artifact is worth naming too: because USDA rounds calories to whole numbers, any food listing 1 or 2 calories per 100 grams produces a wildly unstable ratio. This is why brewed green tea appears to deliver 22.0 grams of protein per 100 calories, and why a sugar-free energy drink outranks Alaska pollock. Neither is true. Neither is exactly false. Both are what happens when you divide by a rounded number near zero. Treat everything above roughly 20 grams per 100 calories with one eyebrow raised, and everything under 5 calories per 100 grams as noise.
What actually sits at the top
With the division artifacts set aside, the ranking opens with soy protein isolate at 27.5 and 26.4 — two preparations of the same industrial powder, and not really a food so much as a purified fraction of one. Third is unsweetened dry gelatin at 25.6, which is a genuine curiosity: gelatin is nearly pure protein by weight, but it contains no tryptophan at all and very little of several other essential amino acids, so as a protein source it is technically spectacular and nutritionally incomplete. A ratio cannot tell you that. Only an amino acid table can.
Then the real foods begin, and they are overwhelmingly cold-water fish and Alaska Native subsistence foods: Alaska pollock at 24.5, Pacific cod at 24.4, bearded seal at 24.3, white-winged scoter at 24.0, beluga at 23.9, shrimp at 23.6, dried walrus at 22.7, dried caribou shoulder at 21.9. This was the finding that surprised me most in 2015 and still does. Eighty-one of the first hundred entries are meat or fish, most of them wild northern animals eaten as whole lean muscle, dried or raw. Whatever else can be said about Arctic diets, they are not short on protein per calorie.
My original surprises, revisited
Here is what I wrote the first time, more or less, with what I now know about each.
How proteinaceous native Inuit foods are. This held up entirely; see above. The mechanism is leanness — subsistence cuts are muscle, often dried, without the fat cap that drags a domesticated cut down the list.
Restaurant cole slaw is below fruit pie. Also true, and still funny. KFC's cole slaw sits at rank 7,888 with 0.63 grams per 100 calories; commercially prepared blueberry pie is at 7,758 with 0.78. Cabbage is a fine vegetable. Cabbage suspended in mayonnaise and sugar is a dessert wearing a vegetable's coat.
Fish and meat rank higher than I knew, but not as high as soy. Correct as stated, though the soy in question is isolate powder, not tofu or edamame. Whole soy foods do very well — the best silken tofu lands at rank 270 — but they do not beat cod.
Rock on, pumpkin seeds! I have to take this one back, and I enjoy taking it back. The entries I saw near the top were SUNFISH, PUMPKIN SD, which is not a pumpkin seed. It is the pumpkinseed sunfish, Lepomis gibbosus, a small freshwater panfish, sitting at rank 53 with 21.8. Actual pumpkin and squash seed kernels are at rank 3,731 with 5.4 — good, respectable, nowhere near the top, and heavy with oil. Abbreviated USDA descriptions are a reliable source of this kind of mistake, and I have left the descriptions verbatim below so you can catch mine.
Rock on, spirulina! This one survives intact. Raw spirulina is at rank 20 with 22.8, the highest-ranking plant-group food on the entire list, and the only one to keep company with the fish. Dried spirulina is at 207. After it, the best vegetable is raw watercress at rank 115 with 20.9, then the leafy tips of bitter gourd, then pumpkin leaves, then frozen asparagus. Leaves, in general, do far better than people expect; it is simply that no one eats 100 calories of watercress, which would be roughly three pounds of it.
What is "raw cisco" anyway? A freshwater whitefish of the Great Lakes and northern Canada, Coregonus artedi, also called lake herring. Rank 240, at 19.4. Smoked, it drops to 2,438.
I don't even want to think about "raw, turtle, green". Rank 31, at 22.3. I still don't.
The bottom, and what it is made of
Three hundred and forty-two entries round to zero grams of protein per 100 calories. They are, almost without exception, the same four things: oils, sugars and syrups, hard candy, and fruit juice. Corn oil, cane syrup, grenadine, sugarless gum, canned pear juice for babies. This is the least surprising finding in the whole exercise and, I think, the most quietly damning one — not because these foods are wicked, but because of how many of the 8,789 rows are versions of them. Scroll the last two thousand entries some time. The American food supply, as catalogued by its own government, has an enormous tail.
What a ratio cannot see
I want to say one thing from the other side of my desk. Ayurveda has almost no interest in this number. It would want to know whether the food is heavy or light, warming or cooling, whether it kindles agni or smothers it, and — above all — whether this person can actually digest it. Dried seal meat is magnificent by protein-per-calorie and would be a poor supper for most people reading this. Raw spirulina is a triumph on paper and cold, drying, and faintly oceanic in the mouth. Gelatin is 25.6 and missing an essential amino acid.
A ratio is a single question asked very quickly of a great many things. It is genuinely useful — I built the list because I wanted it, and I still consult it — but it answers one question and then falls silent. The silence is where the rest of nutrition lives.
Download the data
All three files carry the same 8,789 rows, sorted from most protein per calorie to least. The spreadsheet is the original 2015 file; the PDF is the same list typeset for printing across 189 pages; the JSON is what powers the searchable table below, and is the easiest to work with in code.
The underlying source is the USDA National Nutrient Database for Standard Reference, which in 2015 lived at ars.usda.gov/Services/docs.htm?docid=8964 and now lives, in its successor form, at FoodData Central. USDA data is a work of the United States government and is free of copyright. The protein-to-calorie ratios, the category assignments, and every opinion above are mine.