This article describes a preliminary study of screening/diagnostic instruments for prediction for large-scale application in the military field at the Neuropsychiatric Department of the Military Hospital of Legal Medicine of Verona and for the prevention of self-destructive behaviors, particularly through the use of drugs. 170 subjects divided into three subsamples were examined. The first subsample was characterized by a strong tendency towards normalcy, the second by a strong tendency towards pathology, and the third by a great variety of expressions of psychological and social problems, which were not necessarily related to drug use. These subjects were administered a questionnaire designed according to Squashing Theory principles (Buscema, 1994a). Answers were processed by an Artificial Neural Network created by Semeion in Rome (Buscema, 1996) and were compared with a standard clinical psychiatric assessment report and with the results of psychodiagnostic tests. Results document ANNs' remarkable ability to recognize subjects with declared, in exordium and "at risk" pathological behaviors. Blind results on learning and trial samples show a very high predictive capacity (over 90%). A comparison with the examined subjects' clinical report and the results of the first follow-up also document very high agreements. The broad variation of answers obtained in the third subsample allows further methodological reflections on the contribution of Artificial Neural Networks and Squashing Theory to the study of deviance, for both sociologists and clinicians, and not only for those in the field of drug addiction.