Sixty-four fifth and sixth-grade pupils were taught number series extrapolation by either an algorithm, fully prescribed problem-solving method or a heuristic, less prescribed method. The trained problems were within categories of two degrees of complexity. There were 16 subjects in each cell of the 2 by 2 design used. Aptitude Treatment Interaction (ATI) was an object of the research, with short- and long-term instructional effects investigated. Training was given by computer-assisted instruction. Results revealed an over-all superiority of the algorithm programs for both training levels. An ATI was shown between Numerical Ability and the treatment factor only for the less complex problem category. A reverse relation holds for Negative Fear of Failure and treatment, significant in only the more complex problem category. Results of testing done some days after the training session showed opposite overall effects. No ATI's were found here. Further, Field Dependency appeared to be a good predictor of the susceptibility for set fixedness to the trained problem-solving method. (MP)