The limit of detection is the lowest analyte concentration that a procedure can reliably distinguish from a blank, but not necessarily measure with acceptable accuracy. It is a property of the whole procedure — instrument, matrix, preparation and operator — not of the instrument alone. ICH Q2(R2) expresses it as LOD = 3.3σ/S.
Most disputes about detection limits are not arithmetic disputes. They are disagreements about what a number at or below the limit licenses you to claim. This page covers the conceptual ladder from blank to quantitation, the LOD/LOQ distinction, and how to report a result that falls underneath it. The calculation routes themselves are covered on the two supporting pages linked below.
What is the limit of detection?
IUPAC defines the detection limit as the concentration derived from the smallest measure that can be detected with reasonable certainty for a given analytical procedure. Two parts of that definition do the work.
“Reasonable certainty” means the limit is a statistical statement, not a hardware specification. It encodes an agreed tolerance for being wrong in both directions — calling a blank a detection, and missing a real analyte. The conventional 3.3 multiplier holds both error rates near 5%.
“For a given analytical procedure” means the number belongs to the method as executed, in the matrix it was validated in. The same instrument running the same analyte in plasma rather than in mobile phase will not have the same detection limit. An LOD quoted without its matrix and procedure is not a usable figure.
Chromatography Troubleshooting Decision Engine
Any HPLC symptom, one starting point — the engine narrows hundreds of failure modes to the few that fit your evidence.
LOD vs LOQ: what is the difference?
The limit of detection answers a yes-or-no question; the limit of quantitation answers a how-much question. They are set by different multipliers because they control different things — detection risk in one case, measurement precision in the other.
| Limit of blank (LOB) | Limit of detection (LOD) | Limit of quantitation (LOQ) | |
|---|---|---|---|
| Question it answers | What can a blank plausibly read? | Is the analyte present? | How much analyte is present? |
| ICH Q2(R2) formula | Not defined in ICH Q2 | 3.3σ/S | 10σ/S |
| Signal-to-noise criterion | — | ≈3:1 | ≥10:1 |
| What it controls | False-positive rate from blanks | False positives and false negatives, each ≈5% | Precision — typically ≈10% RSD at the limit |
| What you may claim | Nothing about the sample | Analyte detected; the value itself is not reportable | A numerical result, with stated accuracy and precision |
| Confirmation expected | — | Detected in all replicates at the limit | Meets accuracy and precision criteria at the limit |
The practical consequence: a peak between LOD and LOQ is real but not reportable as a number. You may state that the analyte was detected. You may not state that it was present at 0.8 ppb, because the procedure has not demonstrated it can measure 0.8 ppb with defensible precision. Reporting a value from that band as though it were quantitative is one of the more common findings in method-validation review.
What is the limit of blank, and how does it relate to LOD?
The limit of blank is the highest apparent concentration expected when blank replicates are analysed — the upper edge of blank behaviour. It is not part of ICH Q2(R2), but it is defined in CLSI EP17, the framework used widely in clinical and diagnostic method validation, and it clarifies what the LOD is measured from:
LOB = mean(blank) + 1.645 × SD(blank)
LOD = LOB + 1.645 × SD(low-concentration sample)The two intervals control different errors: the first the false-positive rate from blanks, the second the false-negative rate at a genuinely low concentration. Where the two standard deviations happen to be equal they sum to 3.29, which is the origin of the 3.3 in the ICH expression — the derivation is worked through on LOD calculation.
They are usually not equal. The reason the second interval uses the standard deviation of a low-concentration sample rather than the blank is that variability usually grows with concentration. Assuming blank variability holds at the detection limit is optimistic, and it is why LODs calculated purely from blanks sometimes fail confirmation.
What does a result below the detection limit actually mean?
It means the procedure could not distinguish the sample from a blank. It does not mean the analyte is absent.
This distinction carries real regulatory weight. A pesticide residue below an LOD of 10 µg/kg is not evidence of a residue-free sample; it is evidence that if a residue is present, it is below 10 µg/kg. Where a specification is set near the detection limit, a “not detected” result is only as meaningful as the limit behind it — which is why detection limits, not just results, belong in the report.
The same logic applies to a method whose LOD sits above the regulatory threshold it is meant to police. Such a method cannot demonstrate compliance no matter how many clean results it produces. Checking the LOD against the specification limit is a method-selection step, not a validation afterthought.
How do you report a result below the detection limit?
Report it as censored, not as a number. Conventional forms:
- <LOD, with the numerical limit stated — the least ambiguous option
- ND (not detected), again with the limit stated; without the limit, “ND” is uninterpretable
- <LOQ for results between LOD and LOQ, where detection is confirmed but quantitation is not
Where below-limit results feed a statistical summary — a mean exposure, a batch trend — substitution is common: replacing each censored value with zero, with LOD/2, or with LOD/√2. All three introduce bias, and the bias grows with the proportion of censored data. A common working rule treats substitution as tolerable below roughly 15% censoring and unreliable above it, where maximum-likelihood or regression-on-order-statistics methods are the sounder choice — though the statistical literature on censored environmental data, Helsel in particular, argues against substitution at any censoring rate. Whatever you use, state it: an unlabelled mean over censored data is not reproducible.
Never report a negative concentration. Blank-subtracted responses near zero legitimately go negative, and the arithmetic is not wrong, but a negative concentration is not a physical result. Censor it.
Analytical sensitivity vs functional sensitivity: what is the difference?
Two terms that get used interchangeably and should not be.
Analytical sensitivity is, strictly, the slope of the calibration curve — the change in response per unit change in concentration. It is the S in 3.3σ/S. A steeper slope improves the detection limit for a given noise level, which is why “sensitivity” gets used loosely to mean “low detection limit”, but the two are not the same quantity.
Functional sensitivity is the lowest concentration measurable with a defined imprecision — conventionally the concentration at which the coefficient of variation reaches 20%. It is used mainly in clinical and immunoassay contexts, and it typically lands above the ICH LOQ, because 20% CV is a looser criterion than the roughly 10% relative standard deviation a response of 10σ implies. ICH Q2(R2) states the 10× multiplier without stating that rationale, so treat the correspondence as a convention rather than a definition.
A method can have excellent analytical sensitivity and poor functional sensitivity: a steep slope buys nothing if replicate imprecision at the low end is high. When a supplier quotes a “sensitivity” figure, establish which of the three quantities they mean before comparing it to your own.
How is the limit of detection determined?
ICH Q2(R2) recognises four approaches: visual evaluation, signal-to-noise, the standard-deviation-and-slope calculation, and demonstration of accuracy and precision at the lower range limits. None is mandated. What reviewers look for is that you state which you used and confirm the result experimentally.
Detailed procedures for each, including a worked calculation and Excel functions:
- LOD calculation — blank standard deviation, calibration curve, and the ICH Q2(R2) formula
- Signal-to-noise ratio determination — the compendial S/N formulas and their measurement pitfalls
Note that the EPA maintains a separate construct, the method detection limit (MDL), defined in 40 CFR Part 136 Appendix B as t(n−1, 0.99) × s over spiked replicates, with a paired blank-based component. It controls false positives at 1% and says nothing about false negatives, so an MDL and an ICH LOD on the same method are not interchangeable figures. Environmental laboratories reporting against both should expect them to differ.
Does a lower detection limit always mean a better method?
No, and treating it as a figure of merit to be maximised causes real problems.
A detection limit only needs to be low enough to serve the decision the method supports. If the specification is 10 µg/kg, an LOD of 1 µg/kg is sufficient and an LOD of 0.01 µg/kg buys nothing while typically costing more sample preparation, longer run times, or a more demanding instrument platform.
Pushing the limit lower also has analytical costs. Methods operating close to their detection limit are more sensitive to matrix variation, carryover and contamination, and they generate more results in the awkward band between LOD and LOQ — results that are real, unreportable as numbers, and time-consuming to explain. Aggressive preconcentration can concentrate interferences alongside the analyte, degrading specificity as it improves sensitivity.
The useful question at method design is not “how low can this go” but “is the detection limit comfortably below the concentration at which a decision changes, with margin for matrix variability and instrument drift over the method’s life.”
Why does the same method give different detection limits in different laboratories?
Because the LOD is a property of the procedure in execution, and four things move underneath it.
- Matrix. Co-eluting matrix components raise baseline noise and can suppress or enhance response. A detection limit established in solvent will not survive transfer to plasma, soil extract, or fruit homogenate.
- Instrument state. Detector lamp hours, source cleanliness, column age. Noise drifts upward across a column’s life, and with it the detection limit.
- How σ was estimated. Blank replicates, residual standard deviation and y-intercept scatter capture different variance components and give different numbers on identical data.
- Data processing. Digital smoothing, integration parameters and noise-measurement windows all change the apparent limit without changing the chemistry.
The corollary matters at method transfer: a receiving laboratory should redetermine the detection limit rather than inherit the number in the method document. If your limits move after a transfer, or after a column change, that is expected behaviour — but a sudden shift outside a transfer is worth diagnosing as a sensitivity problem. See the LC-MS sensitivity loss checklist.
Limit of detection: frequently asked questions
Is LOD the same as sensitivity?
No. Sensitivity is strictly the slope of the calibration curve. The detection limit depends on the slope and on the noise, so improving the slope alone does not necessarily improve the limit.
Can the limit of detection be zero?
No. Every procedure has non-zero response variability, so there is always a concentration below which analyte and blank are not separable.
Should LOD be reported in the certificate of analysis?
Wherever a result is reported as not detected, yes. “ND” without an accompanying limit conveys no information about what was ruled out.
What is the difference between LOD and MDL?
LOD as defined in ICH Q2(R2) controls false positives and false negatives at roughly 5% each. The EPA method detection limit controls false positives at 1% and does not address false negatives. They are different statistical constructs and generally give different numbers.
Can a result between LOD and LOQ be reported as a number?
No. Detection may be reported; the value may not, because the procedure has not demonstrated acceptable accuracy and precision at that level. Report as <LOQ.
How often should detection limits be redetermined?
After any change affecting response or noise — detector, column chemistry, sample preparation, matrix — and on method transfer to another laboratory. Many laboratories also verify limits periodically as part of ongoing method performance monitoring.
Setting validation parameters for a new method? The LabVeda AI Method Builder generates validation parameter sets, including LOD and LOQ determination routes, for chromatographic methods. Free, vendor-neutral.
References
- ICH Harmonised Guideline Q2(R2), Validation of Analytical Procedures, adopted 1 November 2023, section 3.2.3.
- IUPAC, Compendium of Chemical Terminology (the Gold Book), “limit of detection”.
- US EPA, 40 CFR Part 136, Appendix B, Definition and Procedure for the Determination of the Method Detection Limit.