Abstract: The aim of this research was to reveal the effect of TGT cooperative learning model with talking chips technique, TGT cooperative learning model, and direct learning model to the mathematics achievement in polyhedron subject viewed from mathematics communication ability. The type of this research was a quasi-experimental by 3x3 factorial design. The population was the eighth grade of Junior High School students at Wonogiri regency on academic year 2013/2014. The sample was taken by using stratified cluster random sampling technique. The instruments used for collecting data were mathematics achievement test and mathematics communication ability test. Analysis of the data used was unbalanced two-way analysis of variance. The conclusion of this research shows that TGT cooperative learning model with talking chips technique gives the same achievement as TGT cooperative learning model, but gives better achievement than direct instruction model, and TGT cooperative learning model gives better achievement than direct instruction model. It shows the same result in each categories of mathematics communication ability. Another conclusion shows that students with high mathematics communication ability have the same achievement as students with middle mathematics communication ability but they have better achievement than students with low mathematics communication ability, and students with middle mathematics communication ability have better achievement than students with low mathematics communication ability. It shows the same result in each learning model.Key words: TGT, Talking Chips, and mathematics communication ability