NIR Spectroscopy Laboratory Guide
A practical guide to NIR spectroscopy in the lab: FT-NIR vs dispersive vs filter instruments, Beer-Lambert law, sample preparation and method development.
Key numbers
How NIR Spectroscopy Works
Near-infrared light (780–2500 nm) interacts with chemical bonds in samples through molecular vibrations. When NIR photons hit a sample, some wavelengths are absorbed and others are reflected or transmitted. The absorbed wavelengths correspond to specific bond types: O-H bonds in water absorb strongly around 1940 nm and 1450 nm; N-H bonds in proteins absorb near 2050 nm; C-H bonds in fats absorb near 1720 nm and 2310 nm.
The instrument's detector measures how much light returns (reflectance) or passes through (transmittance) at each wavelength. This creates a spectrum — a fingerprint of the sample's composition. A calibration model then translates that fingerprint into predicted values.
Beer-Lambert law provides the theoretical foundation: absorbance is proportional to concentration. In practice, NIR spectra are complex overlapping signals that don't follow Beer-Lambert linearly, which is why chemometric methods like PLS regression are required rather than simple single-wavelength calculations.
Near-Infrared Instrument Types
FT-NIR (Fourier Transform)
Highest wavelength accuracy, full spectral range, excellent reproducibility. Preferred for multi-product labs and research.
Higher cost, moving parts require maintenance, sensitive to vibration.
Multi-product food labs, research, method development.
Dispersive (Diode Array)
No moving parts, fast scanning, rugged. Good for industrial environments.
Lower wavelength resolution than FT-NIR. Spectral range may be narrower.
Process environments, at-line industrial measurement.
Filter-Based
Lowest cost, extremely simple, fast. Designed for specific applications.
Fixed wavelength selection — calibrations are locked to the instrument design.
Single-product high-throughput operations (grain elevators, single-parameter QC).
Sample Preparation: Where Most Near-Infrared Errors Start
Sample preparation is the most underappreciated source of NIR error. Instrument precision is often excellent — 0.01–0.05% reproducibility within a session. But that precision is meaningless if the sample in the cup doesn't represent the lot being tested, or if the physical presentation changes between measurements.
Reference Methods: What Near-Infrared Spectroscopy Is Compared Against
NIR is a secondary method. Every calibration is built against a primary reference value — the chemistry that NIR is taught to predict. Knowing the reference method behind your calibration tells you what NIR's accuracy ceiling is, when to expect drift, and which results to defend in an audit. The reference methods below are the ones food, feed, and dairy NIR calibrations are typically built against.
Method codes link to the issuing organization's official catalog. AOAC, AOCS, AACC, ASBC, EBC, and IDF method documents are paywalled — search by code on the catalog page to find the specific method. ISO standards link to the dedicated standard page (free preview, paywalled full text).
Why the reference method bounds NIR accuracy
PLS regression learns the relationship between NIR spectra and the reference Y values you give it. If your Kjeldahl protein values have ±0.15% reproducibility between operators, the NIR cannot predict protein better than ±0.15%. The reference's noise floor is the ceiling for NIR.
This is why "near-infrared is more accurate than wet chemistry" is a category error. This analytical approach is more repeatable (same instrument, same sample, very small variation). Accuracy is bounded by the reference data quality.
Sample handling differs by reference method
Each reference method specifies sample preparation: Kjeldahl uses the whole digested sample; Soxhlet uses a defatted thimble; oven drying uses a thin even layer; ANKOM uses a sealed filter bag. Your NIR sample preparation should match what the reference saw — or at minimum, both should sample the same lot the same way.
Skip this and the calibration learns the disagreement between two different sampling protocols, not the underlying chemistry.
When to run reference vs NIR
- • NIR every batch / every truck — speed and frequency
- • Reference 1 in 20–50 NIR results — calibration validation
- • Reference on every NIR-flagged borderline — defensible decision
- • Reference for any audit, supplier dispute, or label compliance challenge — primary method status
- • Reference quarterly on calibration check samples — drift detection
Go Deeper with Our Near-Infrared Course
Our course covers near-infrared (NIR) fundamentals — Beer-Lambert law, instrument design, sample presentation, and method development — with practical examples built for food and feed professionals.
Free access: 3 resources + preview lesson. Upgrade anytime for full course. No spam.
Go deeper in the course
Enroll — $99 lifetime 14-day refund
Related
Questions
Why does my NIR reading disagree with the wet chemistry lab?
Three usual causes. (1) NIR's accuracy is bounded by the reference method that built its calibration — if Kjeldahl protein has ±0.15% reproducibility between operators, the NIR cannot predict tighter than that. (2) The sample being scanned isn't the same physical preparation the reference saw — Soxhlet uses a defatted residue, NIR scanned the whole ground sample. (3) The lot is outside the calibration's training range. Re-run the wet chemistry on a duplicate aliquot before assuming NIR is wrong.
How often should I run wet chemistry checks against my NIR?
For routine QC, 1 reference assay per 20–50 NIR results plus reference confirmation on every NIR-flagged borderline sample. For audits, supplier disputes, and label-compliance challenges — always run reference. Quarterly, run reference on a fixed set of calibration check samples to catch slow drift. Plants that try to drop wet chem entirely lose calibration confidence within 12–18 months.
Can NIR replace Kjeldahl, Soxhlet, or oven drying for compliance reporting?
For internal QC and process control — yes, in most food/feed/grain matrices. For label compliance, AOAC-defensible audit responses, and supplier disputes — no, the AOAC reference method is the primary defensible result. The right framing is not "replace" but "NIR for frequency, wet chemistry for defensibility." NIR handles every batch; wet chem handles the borderline and the audit.
Which wet chemistry method should be the reference for my NIR calibration?
Match it to the parameter and matrix. Protein in grain/feed: Kjeldahl (AOAC 991.20) or Dumas (AOAC 990.03). Fat in grain/feed: Soxhlet (AOAC 920.39). Fat in dairy: Röse-Gottlieb (IDF 1) or Mojonnier (AOAC 989.05). Moisture in most matrices: air oven (AOAC 925.10). Moisture in low-water powders: Karl Fischer (AOAC 962.16). ADF/NDF: ANKOM filter bag (AOAC 973.18 / ISO 13906). Ash: muffle furnace (AACC 08-01). Reference reproducibility figures are listed in the table above.
How fine should I grind samples to match the wet chemistry reference?
Match the prep procedure the reference method specifies, then keep it consistent for NIR. Most calibrations are built on 1 mm Cyclotec or UDY ground samples; some grain elevators use whole-grain NIR which doesn't need grinding. The NIR sample must be representative of what the reference assay measured — same lot, same homogenization, same particle size distribution. Skip this and the calibration learns the disagreement between two prep protocols, not the chemistry.
What does a daily performance check actually check, and how is it different from validation?
A daily performance check uses a stable internal standard (a sealed reference cup or similar) to confirm the instrument hasn't drifted overnight — wavelength, baseline, and noise. It says nothing about whether the calibration is still accurate against current production samples. Calibration validation is a separate exercise: take real production samples, run NIR and the wet chemistry reference in parallel, plot residuals, decide whether slope/bias adjustment is enough or whether a full calibration update is warranted.