Clinchy and Rosenthal's error classification scheme was applied to test results to determine its ability to differentiate the effectiveness of instruction in two elementary schools. Mathematics retention tests matching the instructional objectives of both schools were constructed to measure the understanding of arithmetic concepts and the ability to perform computations and algorithmic operations. Inter-school comparisons were made with respect to the types of errors made: computational, algorithmic, and omission. Fifth grade students in one school made significantly fewer computational errors than in the other school. In the remaining grades, there were no significant differences. Additional practice exercises were recommended to correct this error. Other students made significantly more algorithmic errors and a four-step approach to teaching algorithms was outlined to alleviate this weakness. Significant inter-school differences in omissions errors were noted in grades 5 and 6--the students concerned would also benefit by a more systematic approach to algorithm teaching. In sum, the error classification scheme can be used to evaluate instructional programs and to suggest instructional improvements. (CP)