The construction sample consisted of 95 female borstal trainees, 45 of whom had injured themselves. All the trainees were at Bullwood Hall Borstal (Great Britain). They were assessed from April to October 1979. Variables examined included historical information, personality questionnaire data, correctional environment information, and the circumstances surrounding the self-injury. The nine variables found to discriminate between the two groups included previous psychiatric treatment, previous institutional experience, previous self-injury, high scores on neuroticism and aggression, low scores on happiness and self-esteem, previous commitment to a borstal, a high psychoticism score (self-injurors only), previous violent offenses, and a low extraversion score (suicide attempters only). Risk scores were computed for 92 additional female offenders at intake. A risk score of 5 or above accurately predicted 83 percent of subjects who subsequently injured themselves. Based on the above results, an intervention strategy based on behavior modification was devised. Intervention included counseling, practical help, and rewards for increasing injury-free periods. A comparison of self-injury for 3-month periods before and after implementation of the intervention showed a highly significant reduction in the frequency of self-injury. A subsequent analysis suggested that a more parsimonius model could be derived. This model is being evaluated. Included are 1 graph, 1 data table, and 16 references.