Examined were the ways peer models affect children's self-efficacy in a cognitive learning context and whether the effects of models vary depending on the sex of the subjects. Subjects were 72 fourth and fifth grade students low in subtraction skills. During pretests subjects indicated the extent to which they thought ability, effort, task characteristics, and luck helped them solve problems correctly; judged their capacity to solve different types of problems; and completed a skills test. Subjects were randomly assigned to one of six experimental conditions: male mastery model, male coping model, female mastery model, female coping model, teacher model, and instructional control (no model). All children in the five model conditions received two, 45-minute treatment sessions over consecutive school days. During the sessions, they observed two videotapes presenting models posing and solving subtraction problems; tapes including peer models differed in sex of model, problem solving behaviors, and verbalizations of achievement beliefs and self-efficacy. Subjects then received 40 minutes of subtraction training and practice for 5 days. Posttests similar to pretests were administered. Results suggest that teachers who incorporate peer models into their classroom instruction may help to promote children's skills and self-efficacy for mastering the skills. Neither type of peer model nor sex of subject significantly affected self-efficacy for learning. Both boys and girls judged themselves more similar to the mastery than to the coping model; for each type of model, boys made higher similarity judgments than girls. (RH)