The mouse, dog, and monkey toxicity data on 30 drugs was retrospectively analyzed in comparison with the actual clinical dose schedules used in man. Animal dose schedules were converted to the human schedule and comparisons were made of the human dose versus the large animal toxic dose low, toxic dose high, and lethal dose, the lethal doses for 10% and 90% of normal mice, and the optimal dose in tumor-bearing mice. If the starting dose in Phase 1 clinical trials had been selected by calculating one-third of the toxic dose low (in mg/sq m) in the most sensitive large animal species, 5 of the 30 drugs would have produced significant toxicity in the first patient. The lethal doses for 10 and 90% of normal mice and the optimal dose in L1210-bearing mice were found to offer good quantitative prediction of human toxicity. Determination of a safe and practical starting dose for Phase 1 studies should take into account not only dog and monkey data but also toxicology data in normal and tumor-bearing mice.