Do you have to send your food to a lab for a nutrition label?

That's the whole answer. The rest of this page is the evidence, the limits, and an honest list of the cases where a lab really is the right call.

What FDA actually requires

There is no regulation anywhere in 21 CFR Part 101 that tells you to send a sample to a lab. FDA's position, stated in Guidance for Industry: Guide for Developing and Using Data Bases for Nutrition Labeling (March 1998), is that:

• The manufacturer owns the accuracy. "FDA's continuing policy since the 1970s assigns the manufacturer the responsibility for assuring the validity of a product label's stated nutrient values."
• The data source is yours to choose. "The source of the data used to calculate nutrition label values is the prerogative of the manufacturer."
• Submitting a database to FDA is optional. "Submission of a data base to FDA for the purpose of nutrition labeling is voluntary."

Be careful with one phrase you will see quoted a lot. 21 CFR 101.9(g)(8) says that "alternatively, compliance with the provisions set forth in paragraphs (g)(1) through (6) of this section may be provided by use of an FDA approved database that has been computed following FDA guideline procedures." That is a narrow, optional route for a database FDA has actually reviewed and approved. It is not a statement that USDA FoodData Central, a commercial database, or the tool you bought is "FDA approved," and nobody should sell it to you that way. The reason ordinary database analysis is acceptable is the guidance language above — the manufacturer picks the data source and owns the accuracy of the result — not (g)(8).

So the obligation is an outcome, not a procedure. Your label has to be right. How you got there is your call — and if FDA ever checks, it checks by testing product off the shelf against the numbers you printed, using the tolerances below.

The advice that says otherwise is out of date

If you've been told you must use a lab, you're not imagining it. That advice is genuinely everywhere, including on resources that small food makers trust most.

Forrager, probably the most-read cottage food resource in the US, currently tells readers: "To get accurate nutrition fact info that meets the federal requirements, you need to have your products tested in a lab, which can get expensive and time-consuming."

That's a well-intentioned sentence written by people who have helped a lot of makers, and the second half is true — lab testing is expensive and time-consuming. But the first half doesn't match FDA's own guidance. Federal requirements govern the accuracy of the value, not the instrument that produced it. Nothing in 21 CFR 101.9 requires analytical testing of your finished product.

If you were quoted $1,609–$1,724 for a full nutritional screen and felt sick about it, that quote is real — that's Medallion Labs' published price per sample, doubled for rush. It's just not compulsory. We break the whole market down in how much a nutrition label costs.

Is a database label legally defensible?

Yes. And it's worth understanding exactly how FDA judges a label, because once you see the mechanism, the fear goes away.

FDA doesn't audit your spreadsheet. Under 21 CFR 101.9(g), it buys your product, analyzes it, and compares the result to what you printed:

The sample, 21 CFR 101.9(g)(2). "The sample for nutrient analysis shall consist of a composite of 12 subsamples (consumer units), taken 1 from each of 12 different randomly chosen shipping cases, to be representative of a lot." So compliance is measured on an averaged composite, not on one unlucky jar.

The two nutrient classes, 21 CFR 101.9(g)(3).

• Class I — nutrients added to fortified or fabricated foods.
• Class II — nutrients that occur naturally in the food.

The floor, 21 CFR 101.9(g)(4). This one has a defined scope. It applies where your label declares a vitamin, mineral, protein, total carbohydrate, dietary fiber, soluble fiber, insoluble fiber, potassium, polyunsaturated fat or monounsaturated fat. For those nutrients:

• If the nutrient is Class I, the composite must be at least equal to the declared value — 100% or more.
• If the nutrient is Class II, the composite must be at least 80% of the declared value. The same provision, 101.9(g)(4)(ii), adds that no regulatory action will be based on a Class II value that falls below the declared amount by an amount less than the analytical method's own variability. That allowance sits inside the Class II rule. It is not a general principle covering every nutrient on the panel.

The ceiling, 21 CFR 101.9(g)(5). For calories, total sugars, added sugars, total fat, saturated fat, trans fat, cholesterol and sodium, the composite must not be greater than 20% in excess of the declared value.

The manufacturing allowance, 21 CFR 101.9(g)(6). Reasonable excesses of vitamins, minerals, protein, total carbohydrate, dietary fiber, soluble fiber, insoluble fiber, sugar alcohols, polyunsaturated fat and monounsaturated fat over labeled amounts — and reasonable deficiencies of calories, total sugars, added sugars, total fat, saturated fat, trans fat, cholesterol and sodium under labeled amounts — are acceptable within current good manufacturing practice.

Read that as a picture and it's a band, not a knife edge. Vitamins, protein and fiber may come in at or above label. Calories, fat, sugar and sodium may come in at or below label, and not more than 20% over.

Two practical consequences:

1. Round conservatively and you have a lot of room. Declare calories, fat, sugar and sodium slightly high, and declare fiber, protein and vitamins slightly low, and normal batch variation stays inside the band.
2. A lab result is a snapshot, not a shield. One tested batch does not guarantee the next one. The tolerance band exists precisely because real food varies batch to batch. A lab certificate does not exempt you from 101.9(g); it just tells you where one sample landed on one day.

Note that "at least equal to" for Class I is stricter than the 80% floor. If you fortify — added vitamin D, added calcium, an added-protein bar — build in real overage. That's the one place a database estimate genuinely needs a margin, and it's the same margin a lab-tested product needs.

How database analysis actually works

A database tool builds your panel from three kinds of data, weighted by your recipe:

1. USDA FoodData Central. The federal reference database. Its Foundation Foods and SR Legacy datasets carry analytically derived nutrient profiles for whole and commodity ingredients — flour, butter, cane sugar, chicken thigh, jalapeño. These numbers came out of labs. Somebody already paid for that chemistry.
2. Branded food data. FoodData Central's Global Branded Food Products Database carries label-declared values for hundreds of thousands of packaged products, which is what you need when your recipe calls for a specific brand of chocolate chips rather than generic "chocolate."
3. Your suppliers' spec sheets. For anything proprietary, the manufacturer's technical data sheet or certificate of analysis is the authoritative source. Ask for it. Good suppliers hand it over without complaint.

The software then scales each ingredient to your batch, applies yield and moisture adjustments, sums the nutrients, divides by the serving size from the RACC table at 21 CFR 101.12, and applies FDA's rounding rules. That last step matters more than people expect — FDA rounding is why 0.4 g of fat prints as 0 g and 47 calories prints as 45.

Where database analysis gets it wrong

Being honest about this is the reason you should trust the rest of the page.

• Moisture loss. Baking, roasting and reducing drive off water. Nutrients concentrate. If the tool assumes raw weights and you don't enter a yield, every per-serving number comes out low. This is the single most common error in DIY labels.
• Fat absorption in frying. A doughnut or a chip picks up frying oil that never appears in your ingredient list. Nobody's spreadsheet knows how much. If you deep fry, weigh the product before and after and account for the gain — or test.
• Nutrient degradation from heat and time. Vitamin C, folate, thiamin and some B vitamins drop during cooking and storage. If you're declaring those vitamins, especially near a claim threshold, calculated values run optimistic.
• Novel and proprietary ingredients. Custom fibers, enzyme-modified starches, botanical extracts, fermented bases. If there's no published composition and the supplier won't give you a spec sheet, there's nothing to calculate from.
• Fermentation and enzymatic change. Sugars get consumed, acids get produced, protein gets broken down. Long ferments move meaningfully away from their starting ingredient profile.
• Natural variability of agricultural inputs. The fat in your beef, the sugar in your fruit, the oil in your nuts all vary by season, cultivar and supplier. A lab test measures one lot of that variability. It doesn't remove it.
• Ingredient loss you didn't record. Marinade that gets drained, brine that gets poured off, batter that stays in the bowl.

Most of these are solvable inside software if you enter your data honestly — record cooked yield weight, not raw ingredient weight. Some of them aren't, and that's the list below.

Use a lab when…

• You have an ingredient with no published nutrient data and no supplier spec sheet.
• You deep fry, or your process changes composition in a way you can't measure — heavy oil pickup, long ferments, deeply reduced sauces.
• You're making a nutrient content claim that sits close to its threshold. "High fiber," "good source of protein," "low sodium," "reduced fat." These are legal assertions with defined cutoffs in 21 CFR Part 101 Subpart D. If your calculated value is within a few percent of the line, testing is cheap insurance against a claim you can't defend.
• You're fortifying. Class I nutrients have to hit 100% of declared, not 80%. Verify your overage at least once.
• An export market, importer or retailer demands a certificate of analysis. Some do, regardless of what FDA thinks. Ask before you assume.
• You've received a warning letter, a retailer complaint, or a legal demand. At that point you need evidence, not an estimate.
• You've reached the scale where a $1,700 test is a rounding error and periodic verification is just good QA.

If none of these apply to you, a lab is a $1,600 way to get a number you can already calculate.

Lab vs. software: the comparison

One note on that cost row, because the numbers you see quoted vary wildly. The $1,609–$1,724 figure is Medallion Labs' published price for a full NLEA panel screen, which is what most people mean by "send it to a lab." A lower $50–$300 per recipe, per revision range also circulates (recipecostcalculator.net); that is a different and much narrower scope of work, not a cheaper version of the same test. Compare quotes on what is actually being analyzed before you compare prices.

The row that decides it for most small brands is cost per revision. Labs and consultants price per label. You price per recipe change. Those two things do not line up when you're still figuring out your product.

If you want the panel built from your recipe, Nutrillius does it in about a minute — $39 one time, or $69/month if you're going to keep changing the formula. Build a label.