This research aims to figure out: 1) the effect of Blended Learning towards the mathematical representation; and 2) the effect of Blended Learning towards the Self-Efficacy of the students. The population of this research is all students of grade VIII at SMP Negeri 2 Purwokerto. This is an experimental research with posttest-only design. The samples are chosen using cluster random sampling technique. Class VIII G is assigned as experiment group (Blended Learning), and class VIII D is chosen as control group (Direct Learning). The result of posttest of mathematics representation with hypothesis test of Mann Whitney shows the level of significance of & alpha; = 0,05, so Ho is rejected. In conclusion, the students' competence of mathematics representation in the learning process with Blended Learning is better than the learning process with direct learning. In addition, the result of questionnaire of self-efficacy with hypothesis test of Mann Whitney shows the level of significance of & alpha; = 0.05, so Ho is accepted. In conclusion, the students' self-efficacy in the learning process with Blended Learning is not any better than the learning process with direct learning. In other words, the Blended Learning is effective for improving the competence of mathematics representation but not for the self-efficacy. The possible factors of obstacles are as follows: 1) students' activeness; 2) less frequency of meetings; 3) learning model and language used by teachers; 4) the students' condition with strong self-efficacy already.