Twenty laboratories worldwide participated in a collaborative trial sponsored by the International Programme on Chemical Safety on the mutagenicity of complex mixtures as expressed in the Salmonella/microsome assay. The U.S. National Institute of Standards and Technology provided homogeneous reference samples of urban air and diesel particles and a coal tar solution to each participating laboratory, along with samples of benzo[a]pyrene and 1-nitropyrene which served as positive controls. Mutagenic potency was characterized by the slope of the initial linear component of the dose-response curve. Analysis of variance revealed significant interlaboratory variation in mutagenic potency, which accounted for 57-96% of the total variance on a logarithmic scale, depending on the sample, strain and activation conditions. Variation among replicate extractions of organic material (required for the air and diesel particles) and among replicate bioassays within the same laboratory was also appreciable. The average potencies for air and diesel particles in laboratories using Soxhlet extracts were not significantly different from those in laboratories using sonication, although there was larger interlaboratory variation for the Soxhlet method. Repeatability (which approximates the coefficient of variation within laboratories) ranged from 18 to 40% for air and diesel particles extracted using sonication, depending on the strain and activation conditions. Repeatability of Soxhlet-extracted air and diesel particles, however, ranged from about 37 to 89% including outliers and from about 11 to 31% excluding outliers. Repeatability of the coal tar sample and the 2 positive controls was in the range 18-34%. Reproducibility (which approximates the coefficient of variation between laboratories) was generally at least twice repeatability, and exceeded 100% for Soxhlet-extracted air and diesel particles, as well as 1-nitropyrene. Reanalysis of the data omitting observations of more than 1500 revertants/plate generally had little effect on these results. Elimination of outlying observations had limited impact, with the exception of Soxhlet-extracted air and diesel particles. In this case, reproducibility of bioassay results was notably improved, due largely to the omission of results for replicate extractions which varied more than 5-fold within one laboratory. Normalization of the log potency slopes for the mixtures by the corresponding slopes for benzo[a]pyrene tended to reduce this variation, although variation was increased after normalization by 1-nitropyrene. Adjustment for the percentage of organic matter extracted from the air and diesel particulate samples had little effect on variation for sonication-extracted particles, whereas variation was reduced for diesel particles and increased for air particles for Soxhlet.