This study aims to determine: (1) the significant influence between learning models on metacognitive abilities, (2) significant effects between learning models on mathematical connections, (3) the interaction between discovery learning and problem based learning on metacognitive abilities, (4) the interaction between discovery learning and problem based learning models on students' mathematical connections. This research is a quasi experimental research. The instruments used were: (1) tests of students 'initial mathematical abilities (2) tests of metacognitive abilities, (3) tests of students' mathematical connections with the subject of Lines and Angles. Inferential data were performed using covariance analysis (MANOVA). The result of hypothesis testing shows that the average metacognitive ability in learning using discovery learning learning models is 77.20 > 75 with a standard deviation of 2.161 and students' mathematical connections obtained are 83.63 > 45 with a standard deviation of 3.480. By using the problem base learning model is 76.76 > 75 with a standard deviation of 2.216 and for the average mathematical connection of the students obtained 81.84 > 45 with a standard deviation of 2.096. The data show that the metacognitive abilities and mathematical connections of students who are taught using discovery learning models are better than those who are taught using problem-based learning models