The purpose of this research was to determine the upgrade mathematical representations of college students using learning based theory Valsiner better than the ability of mathematical representations of students who use conventional learning models and know the rest of the indicator to which students have difficulty in solving problems ability mathematical representations after being given learning based Theory Valsiner. The method used is a quasi-experimental. The population in this study are prospective teachers of mathematics education in universities Siliwangi Tasikmalaya. While the subject of the sample were college students who follow courses in semester V probability theory which consists of an experimental class and the control class in Mathematics Education. Samples were taken at random based on class, grade VB as experimental class amount 34 students, and the class as a class VD control with the amount of 34 students. The technique of collecting data using mathematical representations ability tests. The research instrument set of test questions. Data were analyzed using two different test average. From processing data analysis and hypothesis testing using real level test a = 1% result increased ability mathematical representations of students by using learning based on theory Valsiner better than the ability of mathematical representations of students who use conventional learning models, and the greatest difficulties experienced by students in solving ability representation mathematics is the visual representation indicator with a percentage of 39.71%.