In a quality lab, you don’t just want a number you want to know how trustworthy that number is. Relative standard deviation equivalent to the coefficient of variation is the workhorse statistic for answering that question. See our RSD formula and applications sections for the full overview.
The validation question
When you validate an analytical method, you’re asking: if I run this assay ten times on the same sample, how close together will my answers be?
RSD gives you a single, scale-free number that summarizes that closeness. The International Council for Harmonisation (ICH) Q2 guideline is the global reference for these calculations, and the U.S. FDA Analytical Procedures and Methods Validation document codifies the same expectations for U.S. submissions.
A worked example
Suppose you run a calibration sample through your HPLC seven times and read off these peak areas:
98.1, 99.2, 97.8, 98.5, 98.9, 99.0, 98.3
- Mean: 98.54
- Sample SD: 0.513
- RSD: 0.52%
About half a percent. That is good precision, comfortably inside the 2.0% limit that many monographs set for replicate standard injections. The acceptance limit for a specific method still comes from its monograph or validated procedure, as explained in RSD in HPLC. The NIST/SEMATECH e-Handbook of Statistical Methods gives a rigorous treatment of the same calculation.
Setting acceptance criteria
Different industries set different bars. Some commonly cited examples, each with its own context:
- Pharmaceutical assays: many monographs and in-house procedures specify about 2% RSD for repeatability. For assays without a stated requirement, the harmonized USP <621> and Ph. Eur. 2.2.46 text calculates a limit from the assay range and the number of injections, which can be tighter.
- Bioanalytical methods: CV ≤ 15% for quality control samples, and ≤ 20% at the lower limit of quantification, under ICH M10.
- Chromatographic injection repeatability: a 1994 FDA reviewer guidance describes RSD ≤ 1% for five or more injections as desirable.
These are context-specific figures, not universal rules. Your specific method’s regulatory or internal limits override any rule of thumb. ICH Q2(R2) itself sets no numeric precision limits; it defines how precision is studied, from repeatability through intermediate precision and reproducibility. Clinical labs additionally screen results against Westgard rules; manufacturing teams plug RSD into Six Sigma Cpk studies.
What RSD doesn’t tell you
RSD measures precision (closeness of repeat measurements), not accuracy (closeness to the true value). You can have a tightly clustered set of readings that are all consistently wrong. Good QC programs measure both RSD plus recovery against a reference standard. See our Limitations and When not to use RSD sections for the boundary cases.
Related reading
- What is Relative Standard Deviation? the conceptual primer.
- Sample vs Population Standard Deviation which divisor to use.
- Horwitz Equation and HorRat predicted between-laboratory precision by concentration.
- Pooled RSD combining precision estimates from several runs or levels.
- U.S. EPA Guidance for Data Quality Assessment (QA/G-9S) environmental QC framework.
Try a real validation set
Paste your replicate readings into the RSD Calculator and switch to the Sample (n − 1) option. You’ll see RSD, the underlying SD, and the mean side-by-side.