This study is a quasi experimental research that aims to determine whether there is or not differences in mathematical communication skills between students who apply cooperative learning model think-pair-share with students applying conventional learning and to know how to improve students' mathematical communication skills using cooperative learning model Think -Pair-Share. Data collection techniques in this study are observation, test  and documentation techniques. The independent variable in this research is cooperative learning of think-pair-share model, while the dependent variable is student's mathematical communication ability. The data analysis technique used is test-t to see whether there is or not the difference and test of improvement N-Gain to see the improvement of students' mathematical communication ability. Based on the results of the analysis obtained the research results that there are differences in the ability of mathematical communication between students who use cooperative learning model think-pair-share (experiment class) with students using conventional learning (control class). This can be seen from the value of t0 obtained is 2.085 greater than the value of ttable is 1.99. Supported by the average grade obtained by experimental class that is 66,375 higher than the average value of control class is 53,875. Cooperative learning of think-pair-share model can also improve students' mathematical communication ability which is seen based on analysis result with N-Gain improvement test ie experimental class using cooperative learning with Think-Pair-Share model has increased by 0.607 higher than the increase obtained Control class is 0.432.