This study compares self-instructional (SI) and direct training (DT) effects on task performance of impulsive kindergarten children. Fifteen subjects with a mean age of 5.87 years and mean WPPSI IQ of 87.6 were randomly assigned to three groups: SI, DT and control. A pre-test, treatments, post-tests design which utilized Kagan's (1966) Matching Familiar Figures (MFF) Test, the Porteus (1965) Maze Test, and several sets of task items by Cattell, was employed. SI training was designed to teach subjects to use self-instructional verbalizations to guide their performance. The DT group received procedural instructions. The attention-control group practiced treatment group task materials for an equivalent time. There were two training sessions. One-way analysis of variance shows that training effects were significantly different from control effects in the second training session. No significant effects were found among MFF test scores across the pre- and post-testing sessions. Analysis of variance indicates the absence of generalization of training to performance on Raven's (1956) Coloured Progressive Matrices and the classification subtest from Cattell's (1950) Culture Fair Intelligence Test. Verbalizations of all subjects during testing were tape-recorded. No group differences in self-instructional verbalizations were found. Relations between verbalizations and test performance were low and nonsignificant. It is concluded that self-instructional training to improve the task performance of impulsive children should be reassessed. (RH)