Until recently, problem solvers have typically used single-technique-based tools to build the solution. Also in the field of predictive toxicology, a few systems have been developed in that way, with positive preliminary results. One approach to deal with real complex systems is to use two or more techniques in order to combine their different strenghts and overcome each other’s weakness to generate hybrid solutions. In this project we pointed out the needs of an improved system in toxicology prediction. An architecture able to satisfy these needs has been developed. The main tools we integrated are rules, ANN, graph search, and rule learning algorithms. We defined fragments responsible for carcinogenicity according to human experts, developing a module able to recognize these fragments into a given chemical. To each fragment a carcinogenicity category was associated. Furthermore, we developed an ANN, using molecular descriptors as input, to predict carcinogenicity as a real value. PCA was used to reduce the number of descriptors used by the ANN. Finally, we developed an automatic learning program to combine the results obtained from the two previous modules into a single predictive class of carcinogenicity to man. We tuned the system to maximize the predictive power of the system.