The major purpose of this study was to explore the effects of cognitive apprenticeship approach while applied to teaching mathematical problem solving at the elementary school level in Taiwan.Three sixth-grade classrooms at Nantze Elementary School in Kaoshiung were randomly assigned to three different treatments in teaching mathematical problem solving: cognitive apprenticeship group, regular group, and control group. These sample students were tested by Mathematical Problem-Solving Test and Mathematical Problem-Solving Transfer Test. The above tests were implemented three times: pre-treatment, immediate post-treatment, and eight weeks after treatments.Results indicate that the cognitive apprenticeship approach can significantly improve students' problem-solving performance on the curriculum-related problems and transfer problems, and these above teaching effects can last eight weeks. Nearly all interactions between treatments and student's, mathematical levels were not significant except students' performance on understanding problems. There were positive relationships between students' mathematical level and all of the dependent variables.