AbstractThe results of 89 new field trials and 11 supervised trials were considered, together with 91 sets of residue data evaluated earlier. The datasets consisted of 22 643 valid residue data. As all variability factors calculated from individual sample sets are affected by the uncertainties of sampling and analysis, the average of the P0.975/Rave (97.5th percentile of the residue population divided by the average residues in the lot) values gives the best estimate for the variability factor. The Harrell–Davis (H–D) method gave an average value of 2.89 for the variability factor for all samples, while the average variability factors obtained from samples derived from the new field and supervised trials were 2.8 and 2.7 with the IUPAC and H–D methods respectively. The number of residue values below the LOQ in a sample set significantly affects the observed variability factors. It was found that datasets containing over 20% non‐quantifiable residues might not reflect the true variability of the residues. Mixing of treated and non‐treated commodities may significantly increase the apparent variability. Consequently, only datasets of known origin and consisting of well‐quantifiable residues should be used for estimation of the variability factor. Samples taken from marketed lots may not represent a single lot, and thus they have limited value in estimating the variability factor. The large number of residue data confirms the applicability of the default variability factor of 3 adopted by the FAO/WHO for deterministic estimation of the acute intake of pesticide residues. Copyright © 2006 Society of Chemical Industry