This research includes quantitative research using the Quasi Experimental Design method to examine the effect of differences in the use of the ECIRR learning model, the PQ4R learning model and the Direct Instruction model on mathematical problem solving abilities in terms of students' self-efficacy. Data collection techniques used are tests, questionnaires and documentation. Sampling using Cluster Random Sampling. The population in this study were seventh grade students of SMP Darul Ulum Sekampung. Class VII.2 as the experimental class 1 using the ECIRR model, class VII.1 as the experimental class 2 using the PQ4R model, and class VII.3 as the control class using the Direct Instruction model with fractional material. Data analysis used two-way ANOVA of unequal cells. The results of this study are (1) there is an effect of differences in the use of the ECIRR learning model, the PQ4R learning model, and the Direct Instruction model on mathematical problem solving abilities, (2) there is an effect of mathematical problem solving abilities between students who have high, medium, and low self-efficacy. , and (3) there is no interaction between the learning model and self-efficacy on students' mathematical problem solving abilities.