An up-and down method for acute toxicity (LD50) testing has been developed and statistically evaluated. Compared with the "classical" procedure, this method permits a major reduction in the number of animals used. In the up-and-down procedure, animals are dosed one at a time. If an animal survives, the dose for the next animal is increased; if it dies, the dose is decreased. A survey of 48 acute toxicity tests in rats showed that the great majority of the animals that ultimately died did so within 1 or 2 days. Because of this, it suffices to observe each animal for 1 or 2 days before dosing the next animal. It is recommended, however, that surviving animals be monitored for delayed death for a total of 7 days. The procedure for estimating the LD50 takes into account all deaths, and may be performed using widely available computer program packages. Testing in females alone is recommended, based on the observation that they were generally more sensitive in the survey of 48 studies; selective follow-up in males may sometimes be indicated. The procedure has been tested, by simulation, on 10 of the survey studies. It produced excellent agreement with the original studies. The 95% confidence interval for the LD50 averaged +/- 32% by the up-and-down method, compared with +/- 15% for conventional studies using 40 to 50 animals. The up-and-down procedure will require only 6 to 10 animals, provided that the initial estimate of the LD50 is within a factor of 2 of the true LD50. The method cannot be recommended for testing materials where deaths beyond 2 days postdosing are the rule.