Techniques for the prediction, measurement, and improvement of student performance were examined in an introductory physics course required for engineering majors. The contributions of this study include (1) the application of a statistical technique for predicting performance, (2) a computer program for training basic problem-solving skills, and (3) evidence for the value of training with complex homework problems. The prediction of performance was calculated using the method of discriminant analysis and data from the student's academic record. Specifically, this method predicted the chance of a satisfactory grade or a risk of failure using the student's grade point average (GPA) at entry to the course and grades in certain preceding technical courses. The technique was successful in predicting outcome of the course for over 70% of the students and provided a baseline of anticipated performance against which the results of an intervention could be measured. Improvement of performance resulted from two intervention techniques that modified the students' out-of-class assignments. The first intervention was Precision Teaching; a modification of homework exercises designed to improve basic skills in problem solving. Evaluation of this intervention indicated that class performance improved substantially; the number of students failing the course dropped to about one-half of that predicted by the discriminant analysis technique and overall class performance improved by almost one letter grade. The second intervention was based on the use of complex, multi-step homework problems designed to discourage "matching" of problems to operational formulae in the text. Evaluation of this intervention indicated that student performance on specific parts of the curriculum improved by 10%-20%. Both df these techniques resulted in significant improvement in performance largely on the final examinations. (C) 1999 American Association of Physics Teachers.