The first step involved collecting from 223 heroin addicts and 322 nonusers a total of 112 variables that were not traditionally connected to drug users' behavior. Different types of ANNs were used to test the system's capability of classifying drug users and nonusers correctly. Semeion also created a special ANN tool to reduce the number of independent variables. The ANN selected for the first experiment was a Supervised Feed Forward Network. This ANN classified 95 percent of the sample accurately. A special sensitivity tool selected 47 of the 112 independent variables as necessary to train the ANN. The second phase of the research tested different types of ANN on the new 47 variables to which kind of ANN was better able to classify the sample. This benchmark included seven ANNs. The third part of the research used a Constraint Satisfaction Network created by Semeion to simulate a dynamic fuzzy map of the drug user's world by determining the fuzzy variables crucial for differentiating between drug users and nonusers. Findings indicated that it is possible to predict or recognize drug vulnerability with a high degree of accuracy, based on variables that contain no information related to drug-related problems. Tables, figures, author biographies and photographs, and 30 references (Author abstract modified)