The so-called "decision tree" for computed programs for data analyses of toxicity studies indicates that, when a set of quantitative data is heterogeneity in variance among groups, then data should be transformed into ranked data to test the significant difference among mean values. However: 1. Hypothetical sets of data indicate that the minimum sample size needed to detect a significant difference among mean values by multiple-comparison test of ranked data is too large when compared with the practical number of animals (e.g. 4 dogs or 10 rats per sex per group) used in toxicity studies. 2. The homogeneity test of variances among groups may provide itself important information on the toxicity of test compound, as well as the test for significant differences among mean values indicates. 3. On the assumption that variances in data for most test items in toxicity studies are homogeneous as far as the test items are not greatly affected by the treatment with test compounds, it seems practical to test the significant difference among mean values by parametric procedures even if the data are heterogeneous in their variances among groups.