This review concentrates on the use of artificial intelligence (AI) sy
stems for the prediction of toxic hazard via the establishment of stru
cture-activity correlations. Methods for the analysis of the structure
and physicochemical properties of molecules referred to include topol
ogical analysis, molecular orbital calculations, input of chemical str
uctures, molecular modelling, cluster analysis and pattern recognition
. Emphasis is placed on the importance of identifying substructural fr
agments of sufficient size and physicochemical specificity to act as t
oxicophores. Procedures for processing structural information include
decision-tree and probabilistic systems, as well as algebraic and rela
ted statistical analyses for obtaining quantitative structure-activity
relationships (QSARs). The principal differences between knowledge-ba
sed and automated rule-induction expert systems, and their utilisation
for predicting the activity of chemicals, are discussed by reference
to the use of several methods, including DEREK, HAZARDEXPERT, COMPACT,
CASE and TOPKAT. It is concluded that these AI expert approaches have
an important role to play in predictive toxicity screening as alterna
tives to animal experiments. Also, knowledge-based expert systems are
being developed further for risk assessment.