The relationship between chemical structure and toxicity was explored through the compilation of a large reference database consisting of over 600 chemical substances tested for a variety of endpoints resulting in over 2900 no-observed-effect levels (NOELs). Each substance in the database was classified into one of three structural classes using a decision tree approach. The resulting cumulative distributions of NOELs for each of the structural classes differed significantly from one another, supporting the contention that chemical structure defines toxicity. The database was used to derive a threshold of acceptable human exposure for each of the structural classes that could be applied in the absence of specific toxicity data on a substance within one of the three structural classes. The human exposure thresholds provide guidance on the degree of testing and evaluation required for substances that lack toxicity data.