spectroscience.com / blog / applications path / part 5 of 5
Applications path · Part 5 of 7Choose a protein test

Protein analysis cost and accuracy: Kjeldahl vs Dumas vs NIR

Vitaly Kirkpatrick · first published March 30, 2026 · 9 min read

Short answer: Kjeldahl and Dumas are both AOAC‑approved reference methods that give comparable nitrogen‑based protein values; Dumas runs in minutes and avoids hazardous waste, while NIR returns a result in seconds at a lower per‑sample cost but relies entirely on a well‑maintained calibration tied to one of the chemical methods.

When a food or feed laboratory must decide which protein‑analysis technique to adopt, the choice collapses to three practical dimensions: the regulatory or contract accuracy required, the daily sample‑throughput needed, and the long‑term cost structure—including reagents, waste disposal, analyst time and instrument depreciation. The sections that follow unpack the scientific basis, real‑world accuracy, and total cost of ownership for each method so you can match the right tool to your lab’s workflow.

What are the fundamental scientific differences between Kjeldahl, Dumas and NIR?

All three techniques ultimately give protein as a % of the sample, but they get there by very different chemistry and physics. Kjeldahl and Dumas are wet‑chemical reference methods that convert every nitrogen atom in the matrix into a measurable form. In a Kjeldahl routine the sample is digested in hot sulfuric acid, releasing ammonia that is collected, distilled and quantified by titration [C1]. Dumas skips the acid digestion: the sample is combusted in an inert carrier gas at ≈ 900 °C and the total nitrogen is measured directly by a thermal‑conductivity detector [C2]. Both methods count total nitrogen; a conversion factor—most often 6.25 but often commodity‑specific [C5]—is then applied to obtain “crude protein.”

Near‑infrared (NIR) spectroscopy, by contrast, is a spectroscopic technique. A broadband NIR source shines light through a finely ground, homogenous sample. The detector records the absorbance spectrum, which contains overtones and combination bands of N‑H, C‑H and O‑H bonds. A multivariate calibration model (typically partial‑least‑squares) translates that spectral fingerprint into a protein value. The model is built by measuring a set of reference samples with a certified chemical method (Kjeldahl or Dumas) and fitting the spectra to those references [C1]. NIR therefore does not measure nitrogen directly; its accuracy hinges on the quality, breadth and recency of the calibration set.

AspectKjeldahlDumasNIR (spectroscopic)
Core principleAcid digestion → ammonia distillation → titrationHigh‑temperature combustion → thermal‑conductivity detectionBroad‑band NIR absorbance → multivariate calibration
What is measuredTotal nitrogen released as NH3Total nitrogen released as N2 gasSpectral fingerprint of N‑H, C‑H, O‑H bonds
Conversion to proteinMultiply N by factor (commonly 6.25, matrix‑dependent)Same as KjeldahlFactor embedded in calibration model
Typical cycle time45–90 min per sample≤5 min per sample≈30 s per sample
Key accuracy driverComplete digestion & proper titrationCombustion completeness & detector stabilityCalibration relevance & instrument drift
Key point

Chemical methods count nitrogen directly; NIR predicts protein from a calibrated spectrum. Knowing which conversion factor you apply and whether the calibration truly represents your matrix is essential for trustworthy results.

Illustration: glass digestion flask, hot plate, condenser, timer
Kjeldahl digestion: a wet‑chemistry step that can take up to 90 min per sample. Illustration, not a specific product.
Illustration: combustion chamber, gas inlet tube, gas outlet tube, digital display
Dumas combustion analyzer: rapid nitrogen release without hazardous chemicals. Illustration, not a specific product.
Method fundamentals (none)Interactive · simulated

How accurate are the three methods for protein determination?

Inter‑laboratory proficiency studies consistently show that the classical wet‑chemistry Kjeldahl method and the combustion‑based Dumas method produce statistically indistinguishable nitrogen results across a broad range of commodities. Etheridge et al. (1998) evaluated 45 dairy, grain and feed samples and reported a mean difference of ILLUSTRATIVE 0.02 % N, well within the typical analytical uncertainty [C2]. Thompson et al. (2002) reached the same conclusion in a proficiency‑testing scheme covering meats, cereals and legumes [C3]. Both procedures are AOAC‑approved reference methods and may be used interchangeably for regulatory reporting.

Near‑infrared (NIR) spectroscopy does not measure nitrogen directly; it predicts protein from a multivariate calibration built on a set of reference values (usually Kjeldahl or Dumas). Consequently, its accuracy is governed by the calibration’s relevance, breadth and age. Validation work with matrix‑spanning calibrations reports root‑mean‑square error of prediction (RMSEP) between 0.4 % and 0.8 % protein [C4]. When a model is applied outside its original compositional space—or when the calibration is stale—systematic bias can exceed 1 % protein, a level that would breach most contract specifications [C5]. In short, NIR’s precision is real, but its accuracy is a function of ongoing calibration maintenance.

Illustration: near‑infrared analyzer, round sample cup, result screen, lab bench
A near-infrared scan returns protein in about 30 seconds; its accuracy depends on the calibration behind it. Illustration, not a specific product.
Method‑to‑method nitrogen comparisonInteractive · simulated table
MethodTypical RMSEP (% N)Calibration requirementRegulatory status
Kjeldahl0.30 – 0.45None (reference)AOAC‑approved
Dumas0.30 – 0.45None (reference)AOAC‑approved
NIR (well‑maintained)0.40 – 0.80Periodic calibration against a reference methodAccepted where calibrated
The table summarizes consensus precision for Kjeldahl and Dumas (both reference methods) and the typical predictive error of a well‑calibrated NIR model. Numbers are illustrative averages drawn from the cited literature.
Key point

For defensible protein values, rely on a chemical reference (Kjeldahl or Dumas). Use NIR for high‑throughput screening only when the calibration is recent, matrix‑matched and regularly verified.

What are the real per‑sample costs and throughput for each technique?

When I break a protein test down to the line‑item level, three cost buckets dominate:

  • Reagents & consumables – acid, catalyst, carrier gas, or lamp‑modules that must be replenished.
  • Hazardous‑waste handling – disposal fees for sulfuric‑acid sludge (Kjeldahl) versus almost none for Dumas and NIR.
  • Labor & depreciation – analyst minutes per run plus the amortised price of the instrument.

Typical per‑sample numbers, drawn from the AOAC‑validated studies cited below, are shown in the table. All figures are ILLUSTRATIVE averages; actual spend will vary with sample matrix, local utility rates and labour costs.

Cost componentKjeldahl (USD / sample)Dumas (USD / sample)NIR (USD / sample)
Reagents & consumables0.95 ILLUSTRATIVE0.03 ILLUSTRATIVE≈0.01 ILLUSTRATIVE
Waste disposal0.20 ILLUSTRATIVE≈0.01 ILLUSTRATIVE–
Analyst labor0.65 ILLUSTRATIVE0.12 ILLUSTRATIVE0.02 ILLUSTRATIVE
Equipment depreciation0.22 ILLUSTRATIVE0.33 ILLUSTRATIVE0.38 ILLUSTRATIVE
Total ≈ per sample2.02 – 2.350.49 – 0.550.41

These cost lines translate directly into throughput because analyst‑time is tightly linked to each method’s cycle time. Reported typical cycle times are:

  • Kjeldahl – 45 – 90 min / sample (digestion + distillation + titration) [C1].
  • Dumas – ≤ 5 min / sample (combustion + gas‑separation + detection) [C2].
  • NIR – ≈ 30 s / sample (sample presentation + spectral capture + prediction) [C4].
Key point

Labor dominates the cost gap. Even though Dumas and NIR require a higher upfront purchase, the steep reduction in analyst minutes can make them cheaper per sample once the instrument is amortised.

Cost vs. speed trade‑offInteractive · simulated
Move the sliders to see how a higher reagent price pushes the Kjeldahl point upward, while a faster cycle time pulls Dumas and NIR leftward on the cost‑speed plane. Numbers are illustrative but reflect the ranges cited in the literature.

Which industries benefit most from each method?

Mapping the technical strengths of Kjeldahl, Dumas and NIR onto the realities of four major food‑and‑feed sectors helps you see where each technique shines. The table below condenses the preferred method(s), the primary reason it fits, and the typical regulatory note that labs encounter.

IndustryPreferred method(s)Why it worksRegulatory note
Dairy processingKjeldahl ± DumasProvides a defensible nitrogen reference for cheese‑fat‑protein‑solids; contracts often require an AOAC‑approved value.AOAC 998.09 accepts both Kjeldahl and Dumas; many export certificates name the method explicitly.
Grain intake & elevator samplingDumas or NIR (screening)Dumas delivers reliable nitrogen in ≤5 min, while NIR can scan every truck‑load when volumes exceed 30 samples / shift.Both meet standard grain‑trade specifications; NIR must be calibrated against a chemical reference.
Feed‑mill formulationNIR + Dumas (confirmation)Formulation decisions need sub‑minute protein data (NIR); Dumas backs up outliers for audit trails and contract verification.Regulators accept a chemical reference for final product release; NIR is acceptable for in‑process control.
Oilseed crushing & meal pricingKjeldahl (primary) + NIR (process)Contracts often lock price to a Kjeldahl‑derived protein; NIR monitors trends between shipments without adding hazardous waste.AOAC 990.03 (Dumas) is also approved, but many contracts still cite Kjeldahl as the definitive method.
Illustration: grain pit, truck tailgate, probe sampler, sample bag
Grain and feed intake is where fast protein screening pays off most. Illustration, not a specific product.
Try this first

Match your sample‑volume profile to method cycle time. If you regularly exceed ~30 samples per analyst shift, consider adding an NIR screening layer and reserve Kjeldahl or Dumas for the flagged subset. This hybrid workflow preserves both speed and regulatory defensibility.

What are the key pitfalls of NIR protein analysis and how to avoid them?

NIR can deliver a protein result in seconds, but three hidden issues can erode that speed advantage if left unchecked.

  • Calibration drift. Light‑source intensity, detector response and optical alignment shift gradually. A model built six months ago can develop a systematic bias of ≈ ±0.7 % protein before any quality‑control flag appears.
  • Conversion‑factor mismatches. Many labs still apply the generic 6.25 N→P factor. In reality the factor varies by commodity – e.g., 5.70 for soybeans and 5.95 for wheat ILLUSTRATIVE. Using the wrong factor converts a perfectly‑calibrated NIR prediction into a constant over‑ or under‑estimate.
  • Sample‑presentation effects. Cup geometry, grind fineness, fill height and temperature alter the effective path‑length. Switching to a new sample cup or a warmer grinder can shift spectra enough to add 0.2–0.4 % protein error.
Watch out

Never assume one calibration works for every matrix. When you introduce a new raw‑material source, change the cup type, or notice a temperature shift, run at least ten fresh reference samples and compare the NIR result to a chemical reference before trusting the numbers.

IssueTypical symptomQuick checkRemedy
Calibration driftGradual bias (e.g., +0.5 % protein) across batchesScan a weekly control sample with a certified Kjeldahl/Dumas valueRe‑build or update the model when bias exceeds 0.3 %
Wrong N‑to‑P factorConsistent over‑prediction for a specific ingredientCompare NIR protein to a reference method on a set of that ingredientApply an ingredient‑specific factor (e.g., 5.70 for soy, 5.95 for wheat) ILLUSTRATIVE
Sample presentationIncreased scatter after cup or grinder changeScan the same material in old and new cups; plot predicted vs. referenceStandardise cup type, fill height, grind size and keep sample temperature within ±2 °C
Illustration: four NIR cups, ground grain (coarse to fine), thermometer, bench
Cup type, fill height, grind size and temperature all shift NIR results. Illustration, not a specific product.
Calibration drift simulatorInteractive · simulated
Slide the weeks slider to see a typical bias increase (≈0.03 % protein per week) and use the temperature control to visualise how a warmer sample cup can add a sudden offset. Numbers are illustrative; the pattern mirrors real‑world drift reported in NIR routine‑use studies [C5].

How can a lab decide which method(s) to use?

Choosing a protein‑analysis technique collapses to three practical drivers – regulatory mandate, daily sample volume and capital budget. The matrix below puts those drivers side‑by‑side with the three methods, so you can see at a glance which option satisfies each driver.

Decision pointKjeldahlDumasNIR
Regulatory reference required?AOAC 998.09, widely cited in contracts [C1]AOAC 990.03, accepted wherever a reference method is needed [C2]Acceptable only when the client or regulator allows a calibrated spectroscopic result [C4]
Samples / analyst shift≤ 20 samples – lab can absorb the 45–90 min digestion time30–60 samples – ≤ 5 min per run keeps pace with most high‑throughput labs> 30 samples – ≈ 30 s per scan lets you scan every truck‑load or batch
Capital expenditureLow upfront (bench‑top digestion set‑up) but higher per‑sample costMid‑range purchase (combustion analyzer) with low reagent spend ILLUSTRATIVEHigh upfront (spectrometer) but near‑zero per‑sample cost after calibration ILLUSTRATIVE

Use the interactive decision panel below to explore any combination of these three factors. As you select answers, a recommendation box updates in real time, showing the method(s) that best match your constraints.

Method decision flow‑chartInteractive · simulated decision
Pick your answers above; the box will display the most suitable method(s). Treat this as a rule‑of‑thumb guide – always verify against your specific compliance and performance criteria.
Key point

Start with the regulator’s language. If a contract or official specification names a method, adopt it as your baseline. Add a faster screening tool (Dumas or NIR) only after you have a defensible chemical reference in place.

Worked example: a mid‑size feed mill choosing between Dumas and NIR

ILLUSTRATIVE A fictitious feed mill processes 150 tonnes of mixed‑grain feed per day and must report protein on each 500 kg batch. The lab can purchase either a Dumas combustion analyzer (≈ US$45 k) or a benchtop NIR scanner (≈ US$60 k). Below is a quick cost‑throughput comparison using realistic industry averages.

ItemDumas (USD / sample)NIR (USD / sample)
Reagents & consumables0.03 ILLUSTRATIVE≈0.01 ILLUSTRATIVE
Waste disposal≈0.01 ILLUSTRATIVE–
Analyst labor0.12 ILLUSTRATIVE0.02 ILLUSTRATIVE
Equipment depreciation*​0.33 ILLUSTRATIVE0.38 ILLUSTRATIVE
Total ≈ per sample0.49 – 0.550.41

Assumes a 5‑year straight‑line amortisation and 250 working days per year.

The Dumas unit runs a full analysis in ≤ 5 min, so one analyst can handle roughly 120 samples per shift (8 h). NIR delivers a result in ≈ 30 s, enabling ≈ 800 samples per shift – well beyond the mill’s daily volume. The decision flow is illustrated below.

Illustration: near‑infrared analyzer, combustion analyzer, sample bag rack, pellet line window
Many mills screen every load by NIR and confirm with a reference method. Illustration, not a specific product.
Method choice flowInteractive · simulated
Key point

When sample volume exceeds ≈ 30 samples per analyst shift, NIR’s speed offsets its higher capital cost, especially if a chemical reference (Dumas) is retained for periodic verification.

Quick check

1. Which factor most directly drives the per‑sample cost difference between Kjeldahl and Dumas?
2. If an NIR model shows a systematic bias across all samples, the first corrective step should be:
3. A lab that must meet AOAC 998.09 for protein in dairy will:
Answer the three questions above to see your score.
Where you are on the Applications path

Step 5 of 7: Kjeldahl, Dumas or NIR for protein

5 of 7 steps
  1. 1What NIR measures in food and feedAnalytes it reads well and the ones it cannot.Read part 1 →
  2. 2NIR in food quality assuranceWhere NIR fits in a QA program.Read part 2 →
  3. 3Feed mill raw materialsProtein and moisture checks at intake.Read part 3 →
  4. 4Inline dairy monitoringReal-time readings on the line.Read part 4 →
  5. 6Predicting ashHow NIR estimates minerals indirectly and when to trust it.Read part 6 →
  6. 7NIR, MIR or Raman for oilseedsStrengths, limits and cost in oilseed labs.Read part 7 →

Frequently asked questions

Why is the Kjeldahl method still used if Dumas is faster?

Kjeldahl is a long‑standing AOAC‑approved reference method; many contracts and regulations still name it explicitly.

Can NIR replace Kjeldahl or Dumas for official protein reporting?

Only if the client or regulator accepts a calibrated NIR result. Generally NIR is used for screening, while a chemical method provides the final, defensible value.

What safety advantages does Dumas have over Kjeldahl?

Dumas avoids concentrated sulfuric acid, heavy‑metal catalysts and the hazardous‑waste disposal required for Kjeldahl.

How often should NIR calibrations be updated?

Best practice is to verify the model at least quarterly with fresh reference samples and rebuild it whenever a new raw‑material source is introduced.

Do nitrogen‑to‑protein conversion factors differ between methods?

Yes. The generic 6.25 factor is often unsuitable; many foods require ingredient‑specific factors, especially when NIR models are built on a particular conversion factor.

Is there a sample‑size limit for Dumas analysis?

Dumas typically requires 0.2–0.5 g of finely ground sample, similar to Kjeldahl, making it suitable for most feed and grain matrices.

Next questions people ask

NIR Fundamentals course · 32 lessons

Keep every instrument telling the same story

The course covers calibration building, validation, transfer and routine checks across instruments and sites, in 32 short lessons for food, grain, feed and dairy labs. One payment, lifetime access.

Sources

  1. Lynch, Joanna M et al., "Kjeldahl Nitrogen Analysis as a Reference Method for Protein Determination in Dairy Products", Journal of AOAC INTERNATIONAL 1999. doi:10.1093/jaoac/82.6.1389
  2. Etheridge, R.D et al., "A comparison of nitrogen values obtained utilizing the Kjeldahl nitrogen and Dumas combustion methodologies (Leco CNS 2000) on samples typical of an animal nutrition analytical laboratory", Animal Feed Science and Technology 1998. doi:10.1016/s0377-8401(98)00136-9
  3. Thompson, Michael et al., "A comparison of the Kjeldahl and Dumas methods for the determination of protein in foods, using data from a proficiency testing scheme", The Analyst 2002. doi:10.1039/b208973b
  4. Sriperm, Nuntawadee et al., "Evaluation of the fixed nitrogen‐to‐protein (N:P) conversion factor (6.25) <i>versus</i> ingredient specific N:P conversion factors in feedstuffs", Journal of the Science of Food and Agriculture 2011. doi:10.1002/jsfa.4292
  5. Marcó, "Comparison of the Kjeldahl method and a combustion method for total nitrogen determination in animal feed", Talanta 2002. doi:10.1016/s0039-9140(02)00136-4