This research aimed to find out: (1) which one providing better mathematics learning outcomes, GI model or direct learning model, (2) which one providing better mathematics learning outcomes, the high, medium or low learning independence students, (3) interaction between learning models and learning independence on mathematics learning outcomes. This research used the quasi-experimental research method with 2×3 factorial design. The population were all students of the 8 th class of SMP Negeri 6 Surakarta on academic year 2015/2016. The sample was taken using purposive sampling. The instruments used were learning independence questionnaire and mathematics learning outcomes test. The proposed hypothesis of the research were analyzed by using two-way analysis of variance. The conclusions of this research were as follow. (1) GI model provided better mathematics outcomes than direct learning model. (2) The high learning independence student had better mathematics learning outcomes than medium and low learning independence students, and the medium had as good as low learning independence students. (3) there was no interaction between learning models and learning independence on mathematics learning outcomes.