Question-driven technical explainer · Water quality analyzer perspective · Published: October 29, 2026
Table of Contents
Key Takeaways
- The sample lifecycle is usually the culprit, not the instrument. Most interlaboratory disagreements in water testing trace back to collection, preservation, transport and holding time, which is why regulators prescribe a preservative, a container type and a maximum holding time for every parameter [1][2].
- Holding times are shorter than most plants assume. U.S. drinking-water rules require total coliform samples to reach the laboratory within 30 hours, nitrate to be analysed within 48 hours if unpreserved, preserved nitrate plus nitrite within 28 days, and acidified metals samples within 6 months [1][2].
- Method choice alone can split a dataset. In a European Commission interlaboratory comparison on radioactivity in drinking water, only a minority of participants produced results inside the reference range across the board, and one evaluation could not show any participant method to be superior to the rest [3].
- “Not detected” is not a number. The method detection limit is a statistical construct, defined at 99% confidence that the analyte concentration is greater than zero, so two laboratories with different detection and reporting limits can both be right and still print different values [4][5].
- Every result ships with an error bar. ISO 29201 reminds users that expanded uncertainty is obtained by multiplying combined uncertainty by a coverage factor k, typically between 2 and 3. Differences inside that band are statistics, not negligence [6].
- Blind comparisons close the argument. Identical blind samples, an agreed method and reporting-limit matrix, and standardized z-score evaluation turn a two-lab dispute into a data exercise with a defensible verdict.
Two Reports, One Sample: Ask Three Questions First
When two accredited laboratories test the same water and return different numbers, the instinct is to accuse somebody of incompetence. The productive move is narrower: ask what each report actually claims.
Was the tested quantity defined the same way (total versus dissolved metals, total versus free chlorine)? Was the same analytical method used, or two methods that define the parameter slightly differently? And what uncertainty did each laboratory attach to its value?
Interlaboratory variability is rarely a mystery once those three questions are on the table.
This matters commercially. Discharge permit limits, boiler-feed specifications and reuse agreements are all written as single numbers, while every measurement behind them carries a distribution. A plant that understands where laboratory differences come from can negotiate methods and reporting limits up front instead of relitigating every report, and can decide when an online analyzer is a better arbiter than a fourth round of bottle samples.
The Sample Ages From the Moment You Cap the Bottle
Preservation and Holding Time Are Regulatory, Not Cosmetic
Water is not a stable artifact. Bacteria multiply, chlorine decays, ammonia oxidizes, metals shift between dissolved and particulate forms, and pH drifts as CO2 exchanges with the headspace. Preservation and holding-time rules exist because concentration is a function of time since sampling.
U.S. EPA Region 8 training material for drinking-water systems puts the required windows in plain terms: total coliform 30 hours, nitrate 48 hours unpreserved, total nitrate plus nitrite 28 days preserved, asbestos 48 hours [1].
The federal table behind those numbers is more granular still. Under 40 CFR 141.23, metals samples preserved with nitric acid to pH < 2 may be held for 6 months, cyanide for 14 days, and fluoride for 1 month, each with its prescribed container and preservative [2].
So if one laboratory receives your sample inside the window and the second receives a split of the same bottle three days later, the two reports can honestly disagree. The analyte changed between arrivals. The analysts did nothing wrong.
Make the Chain of Custody Do Its Job
Every defensible comparison needs a traceable record: when the bottle was filled, what preservative was added, how the sample was cooled and transported, and when the laboratory logged it in.
EPA’s sample-integrity guidance is built around exactly this discipline. Hold times are measured from collection to preparation or analysis, and chain-of-custody forms are the evidence that the clock was respected [1].
Before challenging a laboratory’s number, check whether your own logistics gave both laboratories the same starting point. A surprising share of “bad lab” stories dissolves when the two bottles turn out to have travelled different distances at different temperatures.
Same Parameter, Different Method, Different Number
Many water-quality parameters can be measured by more than one legitimate method: spectrophotometric versus titrimetric procedures, different digestion protocols, different membrane incubation temperatures. Each method defines its measurand slightly differently, and each carries its own bias profile in a particular matrix.
The European Commission’s evaluation of its 2012 interlaboratory comparison on gross alpha and beta activity in drinking water is instructive here. Only a minority of participating laboratories reported most of their results inside the reference range, a fraction returned nothing but incompatible results, and none of the methods used by participants could be shown to be superior. Even application of the same method in different laboratories did not guarantee comparable results [3].
The practical consequence: when comparing two laboratories, first confirm both used the same method from the same recognized suite (Standard Methods or EPA-approved methods), on the same fraction (filtered or unfiltered), in the same units. If one laboratory reports COD by a dichromate reference method and the other by a high-temperature digestion kit, the disagreement may be definitional, and no amount of retesting will erase it. Contract language should pin the method down by number and edition, not by parameter name alone.
Why “<0.01 mg/L” and “<0.05 mg/L” Can Be the Same Result
Detection and reporting limits are where most non-chemists get lost.
The method detection limit, as defined in 40 CFR Part 136 Appendix B, is the minimum concentration of a substance that can be measured and reported with 99% confidence that the analyte concentration is greater than zero, determined in a given matrix [4]. It is a statistical construct, not a hard physical boundary, and it depends on the laboratory’s instrument, procedure and recent blank history.
Above the MDL sits the reporting or quantification limit: the lowest level the laboratory is willing to report as a number rather than a flag. The Minnesota Department of Health’s guidance for interpreting laboratory reports distinguishes these tiers explicitly, because a result can be “detected but not quantifiable,” a state different laboratories encode differently on paper [5].
Two laboratories analysing the same low-level sample can therefore print “<0.01” and “<0.05” respectively, and both statements are consistent with the same underlying water.
The fix is administrative. Agree the required reporting limit for every parameter before sampling, and require the laboratory to state both its MDL and its reporting limit on each report. Numbers below limits should be discussed as censored data, not compared digit by digit.
Tolerances: How Close Is “the Same”?
ISO 29201, which deals with the variability of test results and the uncertainty of measurement in microbiological enumeration, formalizes what experienced analysts already know: the expanded uncertainty U is the combined uncertainty multiplied by a coverage factor k, typically 2 to 3, so that the interval around the result covers a large fraction of the expected distribution of values [6].
In plain language, a result of 1.0 with an expanded relative uncertainty of ±20% legitimately spans 0.8 to 1.2. That means a 15% difference between two laboratories may be entirely inside the combined uncertainty of the two measurements, especially near a permit limit where concentrations are low and relative uncertainty is largest.
Procurement teams that do this well ask laboratories for their typical expanded uncertainty per parameter and use it to set difference thresholds in advance. “Different” should mean “outside the combined uncertainty bands,” not “any difference at all.”
| Source of variability | Effect on the reported result | How to control it |
|---|---|---|
| Preservation and holding time | Concentration drifts between sampling and analysis; bacteria grow, chlorine and nitrite decay [1][2] | Match bottle, preservative and maximum holding time to each parameter; record collection time and cooler temperature on the chain of custody [1] |
| Method selection | Systematic bias between methods that measure the “same” parameter [3] | Specify the method by name and edition in the contract; request method-comparison data when changing laboratories |
| Detection vs reporting limits | One report shows “<MDL”, the other shows a different flag or a small number [4][5] | Agree a reporting-limit matrix per parameter before sampling; require MDL and reporting limit on every report [5] |
| Matrix interference handling | Recovery bias in complex industrial matrices (colour, turbidity, salinity, reducing agents) | Require matrix spikes and duplicates at an agreed frequency; review recovery values before accepting results |
| Calibration and uncertainty | Different error bars around each value | Ask each laboratory to state expanded uncertainty (k = 2–3) with results [6] |
| Sample heterogeneity | Non-representative aliquots; solids settle between splits | Composite and homogenize; split the sample at the point of collection so both labs start from the same liquid |
Organizing an Interlaboratory Comparison That Produces Answers
A structured comparison is the cleanest way to settle a recurring dispute, and it does not need an external proficiency scheme to be useful.
- Freeze the design. Choose two to four parameters where the discrepancy hurts, for example nitrate, ammonia or residual chlorine. Agree the method, the fraction, the units and the target reporting limit for each, in writing [3][5].
- Prepare identical blind samples. Use a certified reference material diluted into your matrix, or a large homogenized real sample split into identical bottles. Blind means the laboratories do not know which samples are which, or that a comparison is running at all.
- Ship under matched conditions. Same preservative, same cooler temperature, same courier, logged collection and receipt times so both laboratories work inside the same holding window [1].
- Score the deviations. International proficiency-testing practice grades each laboratory’s deviation from the assigned value with a standardized z-score, which normalizes differences across parameters and concentration levels. Deviations outside the acceptable band point to a real problem; deviations inside it confirm the two laboratories agree within the statistics of the methods.
- Close the loop. For any outlier, walk the rows of the table above in order (preservation, method, limits, interferences, uncertainty) before anyone blames an instrument or an analyst.
One comparison rarely settles anything. A quarterly rhythm, tied to your compliance calendar, builds a performance history that makes the next disagreement cheap to resolve.
What This Means for Plants That Live on Data
Every bottle sample passes through the fragile chain described above: preservation chemistry, holding time, courier temperature, two transfers of custody.
An online analyzer deletes most of that chain. The water is measured in process, in real time, by an instrument whose calibration history you control. For parameters where laboratory turnaround shapes decisions (residual chlorine, conductivity, pH, dissolved oxygen, turbidity), continuous measurement does not replace the accredited laboratory. It arbitrates it: when a bottle result looks implausible, the process data stream tells you within minutes whether the plant actually changed.
Shanghai ChiMay builds its online water quality analyzers around that arbitration role, with calibration certificates and test documentation designed to go straight to an auditor (Shanghai ChiMay site and product range). The same discipline applies in-house. Log calibrations, keep the certificates, and your own data becomes as defensible as the laboratory’s.
Interlaboratory comparison work rewards exactly one habit: deciding the rules before the numbers exist. Plants that do it spend meetings improving the process. Plants that do not spend them arguing about the sample.
References
- U.S. EPA Region 8 — Maintaining Sample Integrity and Ensuring Seamless Reporting to EPA (training deck): hold times measured collection-to-analysis; total coliform 30 h, nitrate 48 h unpreserved, nitrate + nitrite 28 d preserved, asbestos 48 h; chain-of-custody discipline. https://www.epa.gov/system/files/documents/2022-04/sample-integrity-and-seamless-reporting-to-epa-region-8.pdf
- 40 CFR 141.23, Inorganic chemical sampling and analytical requirements: parameter-by-parameter table of preservatives, containers and maximum holding times (metals with HNO3 to pH < 2 for 6 months; cyanide 14 days; fluoride 1 month; asbestos 48 hours). https://www.ecfr.gov/on/2025-01-21/title-40/section-141.23
- European Commission Joint Research Centre — Evaluation of the 2012 EC interlaboratory comparison on gross alpha/beta activity concentration in drinking water: participant performance against the reference range and the finding that no participant method could be proven superior. https://publications.jrc.ec.europa.eu/repository/bitstream/JRC104692/science%20for%20policy%20report_evaluation%20of%20the%202012%20ec%20interlaboratory%20comparison%20on%20gross%20alphabeta%20activity%20concentration%20in%20drinking%20water.pdf
- 40 CFR Part 136, Appendix B — Definition and Procedure for the Determination of the Method Detection Limit: MDL defined at 99% confidence that the analyte concentration is greater than zero, in a given matrix. https://www.gpo.gov/fdsys/pkg/CFR-2011-title40-vol23/pdf/CFR-2011-title40-vol23-part136-appB.pdf
- Minnesota Department of Health — How to Interpret Drinking Water Lab Results: method detection limit versus reporting/quantification limits, and why “detected but not quantifiable” appears differently across laboratories. https://www.health.mn.gov/communities/environment/water/docs/com/interpretlabreslt.pdf
- ISO 29201 — Water quality: The variability of test results and the uncertainty of measurement of microbiological enumeration methods: expanded uncertainty U = k·uc with coverage factor k typically 2 to 3. https://dastmardi.ir/wp-content/uploads/2020/11/ISO-29201.pdf
About the author: Written by the Shanghai ChiMay Technical Editorial Team — engineers who specify, calibrate and support Shanghai ChiMay’s online water quality analyzers (pH, conductivity, dissolved oxygen, residual chlorine, turbidity and COD) for municipal and industrial plants, and who regularly help customers reconcile online data with accredited laboratory reports.
