Ninety-eight undergraduate education majors received a battery of ability tests, measuring general reasoning, associative memory, and trait anxiety and were randomly assigned to three groups--no feedback, feedback, and learner control--for a computer-assisted instruction course on Xenograde systems (an imaginary science). Four state anxiety measures were taken: prior to the course, following the administration of stress instruction, at the mid-point, and at the end. Learner control subjects decreased more in state anxiety than those in the feedback condition, while no-feedback subjects remained high throughout. High anxiety subjects made more errors under the feedback condition than under no feedback. The feedback condition facilitated performance for high reasoning subjects, but impaired performance for low reasoning subjects. (Author/RH)