The acute toxicity of chemicals to rainbow trout, as shown by intraperitoneal injections (IP LD50), oral dosing (oral LD50) and aqueous exposure (LC50) was compared with published values for IP LD50S and oral LD50S of mice and rats. The method of comparison was by simple linear regression analyses of log-transformed data, modified to recognize that X (fish toxicity) was neither fixed nor measured without error. Within-species comparisons demonstrated very strong linear correlations (r = 0.866-0.998) between IP and oral LD50S. Variability was least for the fish data since it was all generated in one laboratory. Comparisons between species of IP and oral LD50S gave correlation coefficients ranging from 0.59 to 0.95 with the majority over 0.80. Correlations were best (r = 0.83-0.94) between fish LD50S and rat and mice IP LD50S. Correlations were poorest between fish and mammalian oral LD50S (r = 0.59-0.66) because the sample sizes and the ranges of values were very small. In all cases, the slopes were close to, or equalled, 1.0. Comparisons of fish LC50S to fish or mammalian LD50S were not as successful. Correlation coefficients ranged from 0.19 to 0.83. Presumably the cause was the aqueous exposure. Interactions of the chemicals with water (e.g. dissociation) and with lipid membranes (partitioning) should cause considerable variations in uptake efficiency. However, adjustments of LC50S for dissociation constants and partition coefficients did not improve these correlations, probably because there were few chemicals for which all data were available. These comparisons demonstrate a potential for a wider use of surrogate species in toxicity testing and for adapting existing data from mammalian toxicology to aquatic hazard assessments.