Reproducibility of results was examined in 70 "near-replicate" comparisons consisting of 2 or more long-term carcinogenesis bioassays of the same chemical administered by the same route and using the same sex and strain of rodent. Overall, there was good reproducibility of positivity, target site, and carcinogenic potency in hamsters, mice, and rats. The published authors' opinions about whether the test was positive disagreed in only 9 of the 70 comparisons. Among the 35 comparisons in which all tests of the chemical were positive, 33 of the near-replicates had at least 1 identical target site. The carcinogenic potency values estimated from near-replicate tests in these 35 comparisons were within a factor of 2 of each other in 40% of the comparisons, within a factor of 5 in 80%, and within a factor of 10 in 90%. For the few cases in which the carcinogenic response was not reproduced, analyses suggest two explanations: In mice the discrepant cases tended to have shorter experiment times than average; in both rats and mice the discrepant results tended to be tests of weakly active compounds.