A statistical model has been developed for estimation of acute toxicity. The model, currently operational for rat oral LD50, permits the estimation of rat oral LD50 for untested chemical compounds. Only the chemical structure, partition coefficient, and molecular weight for a compound are needed for estimation purposes. The chemical structure is partitioned into substructural fragments using the CIDS fragment keys. A regression model is developed on the basis of 425 compounds. A test of the regression equation with 100 compounds not used in its design shows that 56 percent of the compounds are predicted with less than 0.4 log unit deviation between extimated and measured LD50. This toxicity estimation model can be readily adapted to other species and to other measures of toxicity by the use of suitable design data bases. The model also identifies the contribution to toxicity of the fragments and physical characteristics. The use of this model can materially reduce the amount of toxicological testing for new compounds. It also permits the ranking of potentially toxic compounds to allow the most likely candidates to be tested. The method may also prove applicable to the determination of optimum dosages for new drugs.