NOVA Groups: Why 'Ultra-Processed' Isn't the Same as 'High-Calorie'

A bowl of steel-cut oats with berries: 400 calories. A bowl of frosted cereal with skim milk: 400 calories. Roughly the same carbs and protein. Similar fiber on paper (though the oats have more, and from a different source).

Are those two breakfasts equivalent? The macro sheet says almost.

The NOVA classification system says not close, and ten years of cohort data are increasingly backing it up.

This is a companion to the nutrition intelligence pillar. That piece covers the full picture — polyphenols, glycemic load, chrono-nutrition, micronutrients, meal photo analysis. This piece goes deep on one specific dimension: how food processing changes outcomes in ways macros do not capture.

---

Where NOVA Came From

NOVA was developed by a research group at the University of Sao Paulo led by Carlos Monteiro, first published in the mid-2000s and refined through the 2010s. The motivating observation was simple: population-level dietary data showed rising rates of cardiometabolic disease, but per-capita calorie intake and macronutrient ratios were not changing in ways that explained the trend. Something else was.

What Monteiro’s group noticed was that the industrial transformation of food — moving from home-prepared meals to factory-formulated products — was happening in parallel. They built a classification to capture that axis, because the existing nutrient-based frameworks could not. A bowl of oats and a bowl of sugary cereal look similar on a macro sheet. They do not look similar when you ask what industrial processes produced them and what additives they contain.

The result is four groups, ordered by processing intensity.

The Four Groups

Group 1 — Unprocessed or minimally processed.

Foods in their natural state or altered only by drying, crushing, grinding, pasteurizing, fermenting (simple fermentation), or freezing. The processing does not add ingredients — it prepares the raw material for storage or consumption.

Examples: fresh fruit, vegetables, legumes, whole grains (including plain oats, brown rice, whole wheat flour), fresh or frozen meat, fish, eggs, plain milk, plain yogurt, nuts without coatings, coffee beans, tea leaves.

Group 2 — Culinary ingredients.

Substances extracted from Group 1 foods, sometimes with simple refining, that are used in small amounts to prepare Group 1 foods. Rarely consumed alone. Culinary salt, sugar, honey, olive oil, butter, vinegar.

Group 3 — Processed foods.

Products made by adding Group 2 ingredients to Group 1 foods and then preserving, modifying, or cooking in ways that could be done at home or in a small bakery. These are recognizable as food.

Examples: bread (made from flour, water, salt, yeast), cheese (milk, salt, cultures, rennet), canned fish in oil, home-style pickles, cured meats made with salt and smoke only, jams made from fruit and sugar.

Group 4 — Ultra-processed.

Industrial formulations of ingredients most of which are not found in home kitchens. The ingredient list typically includes additives — emulsifiers, artificial flavorings, artificial colorings, high-fructose corn syrup, hydrogenated oils, modified starches, preservatives, bulking agents, humectants.

Examples: sugary breakfast cereals, most packaged bread, soft drinks, energy drinks, packaged snacks (chips, cookies, filled wafers), instant noodles, instant soups, frozen dinners, reconstituted meat products (chicken nuggets, hot dogs, most sausages), protein bars with more than a few ingredients, mass-produced ice cream, commercial pastries, flavored yogurts, margarine, infant formulas (as a category), and most products you would describe as “snacks.”

The litmus test is often: could you make this at home? A loaf of bread, yes. A breakfast cereal with eleven ingredients including mono- and diglycerides and tertiary butylhydroquinone, no.

What the Outcome Data Actually Show

Population studies of NOVA Group 4 intake consistently find that higher Group 4 consumption associates with higher incidence of:

• Cardiovascular disease (Srour et al., 2019 in the French NutriNet-Sante cohort; UK Biobank analyses)
• Type 2 diabetes (multiple cohorts)
• Obesity (mechanistic RCTs including Hall et al., 2019 NIH crossover)
• Some cancers (colorectal in particular)
• All-cause mortality (multiple cohorts across different populations)

The Hall et al. RCT deserves a callout because it’s one of the few tightly controlled mechanistic studies. Participants were given two-week isocaloric-offered diets matched for calories, macros, sugar, sodium, fiber, and fat — one ultra-processed, one minimally processed — and allowed to eat ad libitum. On the ultra-processed arm they spontaneously ate about 500 more calories per day and gained weight; on the minimally processed arm they ate less and lost weight. The food was matched on paper. The eating behavior was not.

The mechanistic explanations that are being investigated include:

Matrix disruption. Nutrients delivered without the fiber, cellular structure, and water context they evolved within are absorbed differently. A whole apple and apple juice share sugar and some polyphenols, but the glycemic response and satiety differ substantially.

Additive effects. Emulsifiers like carboxymethylcellulose and polysorbate 80 have been shown in animal and some human models to alter gut microbiome composition and intestinal barrier function. The human evidence is still developing. Artificial sweeteners show similar ambiguity — some data suggesting altered glucose homeostasis and microbiome shifts, no consensus yet on clinical significance.

Hyperpalatability and eating rate. Ultra-processed foods are engineered to be eaten quickly. Faster chewing means reduced satiety signaling, larger meal sizes, and more meals per day. The calorie label is the same; the actual intake is not.

Advertising and food environment. Group 4 foods dominate supermarket shelves, school cafeterias, and quick-service menus. Availability is not a mechanism in the biological sense but it is part of why population intake is what it is.

It is worth being honest about limitations. Most of the outcome data is observational, so causal claims require caution. Residual confounding (by overall dietary pattern, by socioeconomic factors correlated with ultra-processed intake, by activity levels) is a real concern. The Hall RCT is the strongest mechanistic signal but one study. The field is still debating which specific mechanisms dominate. What is clear is that the association is robust across populations, and the magnitude is large enough that it cannot plausibly be explained by calories alone.

Why This Matters for Tracking

A tracker that captures calories, macros, and fiber but not processing level is missing a first-order predictor of long-term outcomes. The practical implementation is not complicated:

Per-item NOVA assignment. Each logged food gets a NOVA group either from a reference database (Open Food Facts has NOVA labels for most packaged products) or from an LLM estimate for unpackaged items. Packaged items are usually easy — the ingredient list tells you. Home-cooked items usually fall in Groups 1 to 3; a homemade curry is not ultra-processed even if it contains many ingredients, because the ingredients themselves are.

Calorie-weighted meal average. Per-meal, compute the calorie-weighted NOVA group: if a meal is 300 calories of NOVA-1 (salmon) and 200 calories of NOVA-4 (a processed side), the weighted average is around 2.2. Per-meal quality indicators can use this as one of three dimensions (alongside protein adequacy and glycemic load).

Daily and weekly Group 4 percentage. Compute the fraction of daily calories from Group 4. A daily NOVA-4 percentage is noisy — one Friday pizza night can swing it. A rolling weekly percentage is the useful number. Most industrialized-country adults are somewhere between 30 and 60 percent. Below 20 is uncommon. Above 70 is a strong candidate for intervention.

Trends, not single days. A three-month downward trend in Group 4 percentage is a genuine dietary shift. A single week of 65 percent is a bad week. Trackers that alarm on single days train users to ignore the metric.

How to Interpret Your Number

The most useful framing for a user is relative and trend-based, not absolute:

• If your weekly NOVA-4 percentage is steady at around 40 percent and your biomarkers (body composition, bloodwork, energy, sleep) are fine, that’s not an emergency. Many populations sit there and do reasonably well.
• If your weekly NOVA-4 percentage is 60+ and climbing, that is worth paying attention to even if nothing else looks off yet, because the outcome data says the trajectory matters.
• If your weekly NOVA-4 percentage is under 25 percent and holding, you are in the minority and you are likely doing most things right from a processing-quality angle. Focus on other dimensions — polyphenol diversity, micronutrient gaps, meal timing.
• A one-week spike during travel or stress is not a signal. A three-month drift upward is.

The other non-obvious lesson from the outcome data is that calorie restriction of ultra-processed food does not fully offset the effect. Eating 1,500 calories of Group 4 is not equivalent to eating 1,500 calories of Group 1 to 3 just because the calorie number is lower. The mechanism research, imperfect as it is, suggests that reducing Group 4 percentage is worth more than reducing Group 4 calories while keeping the fraction high.

Common Misinterpretations

The NOVA literature has been widely misreported. A few corrections worth making explicit:

“NOVA says bread is bad.” NOVA says ultra-processed bread — usually packaged sliced bread with emulsifiers, dough conditioners, and preservatives — is Group 4. Artisan sourdough, or a loaf made from flour, water, salt, and yeast, is Group 3. The distinction is industrial reformulation versus the traditional product.

“NOVA is anti-convenience.” Group 4 is industrial formulations, not quick-to-prepare foods. A frozen bag of vegetables is Group 1. A pre-cooked pouch of plain rice is at worst Group 3. Convenience does not imply ultra-processing. The tracker should not punish microwave meals made from whole-food ingredients.

“Ultra-processed just means high-calorie junk.” Many Group 4 foods are engineered to be low-calorie — diet soda, low-calorie snacks, sugar-free flavored yogurts. The “ultra” is about the industrial process, not the energy density. Some of the most heavily processed products in the supermarket have modest calorie counts.

“NOVA is opinion, not science.” NOVA is a classification system, which is a scientific tool. Whether Group 4 intake is causally responsible for the associated outcomes is the open scientific question. The classification itself is as contested as any — there are legitimate debates about edge cases (is fermented yogurt with a live culture but added sugar Group 3 or 4?), but the framework is well-established in population nutrition research. The Monteiro group, the French NutriNet team, UK Biobank analysts, and the Pan American Health Organization all use it.

What to Do With the Information

Tracking NOVA does not require changing your eating style to “clean eating” or following any specific diet. The outcome data shows that the biggest effects come from reducing the extreme Group 4 percentages, not from pushing toward zero. A diet at 30 percent Group 4 is likely close to what the research rewards. A diet at 15 percent is probably better but with diminishing returns, and the social and practical cost of going lower is often not worth the marginal benefit.

Specific practical moves that generally shift the weekly number by 10+ percentage points:

• Replace packaged breakfast cereal with overnight oats, Greek yogurt and fruit, or eggs
• Replace most soft drinks with water, unsweetened tea, or coffee
• Replace packaged snacks with fruit, nuts, or cheese
• Switch from pre-made sauces and dressings to oil, vinegar, mustard, and salt
• Limit processed meats (hot dogs, most deli meats, commercial sausages) to occasional rather than default

These are not revolutionary suggestions. They are what works in the data.

Back to the Pillar

NOVA is one of seven nutritional dimensions the nutrition intelligence pillar covers. The others — polyphenol diversity, meal-level glycemic load, chrono-nutrition, IARC exposure, 35-nutrient tracking, and the meal photo analysis pipeline that extracts all of this — each deserve the same depth. For the related treatment of how ultra-processed eating and dietary pattern interact, see Your Diet Has a Pattern — Here Are the 8 Common Ones. For the dosing counter the NOVA framework does not address directly, see IARC Group 1 and 2A Carcinogens in Food.

If you want the cross-domain conversation on training and nutrition, the adaptive training intelligence guide is the right next read. If you want the training-nutrition boundary where underfueling lives, the energy availability guide.

For the platform side, Omnio’s food photo analysis assigns NOVA groups automatically during enrichment using Open Food Facts for packaged foods and LLM estimation for unpackaged, and the per-meal quality indicator includes calorie-weighted NOVA as one of its three dimensions alongside protein adequacy and glycemic load.