Data from 38 laboratories using 5 strains of Salmonella typhimurium (TA98, TA100, TA1535, TA1537, and TA1538) were analyzed to determine sources and magnitudes of test data variability. Each laboratory tested the mutagenicity of 4-nitroquinoline-N-oxide by the same protocol, using both its in-house cultures and a set of reference cultures provided to all laboratories. It was found that neither plate-to-plate nor day-to-day variability within a laboratory differed substantially between the in-house and reference cultures for any strain; this indicated no difference in the laboratories' handling of the two cultures. Not surprisingly, on average, plate-to-plate variability was substantially smaller than day-to-day variability within a laboratory, which, in turn, was substantially smaller than inter-laboratory variability. The solvent DMSO was found to have a small (6-7%) but statistically significant depressive effect on the spontaneous mutant frequency for the two plasmid-containing strains, TA98 and TA100, but not for the other three strains. When the mean value and variance of all laboratories for the in-house culture were compared with the corresponding reference culture values for each dose and strain, no major differences were seen. Any increase in mean or variance in the distribution of laboratory means in one of the two cultures could be ascribed largely to a small number of laboratories. Laboratories that reported 'high' or 'low' levels of spontaneous or induced revertants per plate tended to deviate in the same direction for most strains and for both in-house and reference cultures. If 'genetic drift' contributed to the inter-laboratory variability in this collaborative study, it was a minor component that went undetected in our analyses.