The study evaluated the effectiveness of computerized instruction which incorporated errorless discrimination learning principles for teaching functional word discriminations to mentally retarded young children. Three aspects of performance were examined: acquisition of discriminations, transfer beyond the computer, and identification of the discriminative stimuli controlling responding. Two whole-word fading procedures, one with and one without distinctive feature emphasis produced virtually errorless acquisition of male-female and ladies-gents discriminations. While acquisition was superior with the distinctive feature procedure, transfer was more impressive following the other procedure. Probe tests showed that, for 79 percent of the discriminations learned, stimulus control extended beyond the first letter of the word; for 36 percent, control was sufficiently broad to allow children to discriminate very similar words.