The paper discusses applications of both neural networks and discriminant analysis to psychological diagnostics. Features of classification with the aid of canonical discriminant analysis, Bayesian discriminant analysis, discriminant networks, neural networks (with and without optimal topology selection) are considered. Advantages of discriminant networks based on Bayesian discriminant analysis are shown. Statistical criteria for estimating recognition quality are presented, with verification sample sizes being taken into account. Comparisons of output yielded by different supervised learning structures, which detect children's deviant behavior symptoms derived from the Achenbach tests (great sample size) as well as from the inquirer for indirect evaluation of mental development (small sample size), are given.