To investigate the effects of information feedback of learning difficulties to mathematics teachers on students' mathematics achievement, an one-year experiment was conducted in two junior high schools, in which 710 students of grades 7 and 8 were sampled and divided into Experimental Group and Control Group. The teachers of Experimental Group were given information concerning students' learning difficulties before instruction, and were asked to find out learning difficulties by means of unitary diagnostic tests and to remedy them immediately after instruction. The students of Control Group were distributed in other ordinary classes wherein they received regular instruction only. The data collected after one-year experiment were analyzed by Multivariate . Analysis of Covariance, with test scores on Principles and Rules, Calculation, and Application as criterion variables, and with Intelligence, Pre-achievement, Achievement Motivation, Social-economic Status, and Perception-toward-Teachers as control variables. It was found that the overall mathematics achievement of the Experimental Group students was significantly higher than that of the Control Group students. The data were also analyzed by Stepwise Regression Analysis and Canonical Correlation Analysis. Results showed that Pre-achievement and Intelligence were the best two predictor variables of students' mathematics achievement. Accordingly, the authors suggest that (1) more emphasis be put on formative evaluation than on summative evaluation, (2) provide more opportunities for students to master every important subskills in learning hierarchy, and (3) learning be carefully scheduled to match each student's cognitive development level.