The purpose of research in the design of this experiment was to investigate the apparent increase in the mathematical problem solving abilities of students. The research was conducted in the Department of Mathematics Education, State University of Medan. This research is an experimental study with the study design pre-test-post-test control group design. The population in this study were all students of semesters 1 (one) UNIMED mathematics education majors to take samples of two classes (class experimental and control classes) through random sampling technique. Data obtained from tests the ability of solving mathematical problems. Data were analyzed by ANOVA two lanes. Before use ANOVA two-lane first tested for normality and homogeneity in research in this study with a significant level of 5%. The results showed that the average increase problem solving ability test experimental class was 0.48 and 0.38 in the control class is sig = 0.02 with 0.02 <α = 0.05 then there is an increase in mathematical problem solving ability students are taught by learning models IMPROVE higher than in the Direct Learning. The findings of the study recommend IMPROVE be one learning approach used in primary school or university to achieve a high level of competence thinking.Keywords    : Improve Learning Model, Mathematical Problem Solving Ability