An experimental analysis of voting behavior based on a new matrix game paradigm defined by Takigawa (1983) is reported. Five experimental conditions defined by 2×2 payoff matrices were adopted, where the first row represented the player's vote for the first party and the second row the vote for the second party. The first column corresponded to the winning of the first party and the second column to the winning of the second party. Eight successive elections were performed in the course of an hour. The result showed the effectiveness of the payoff matrices used. The selection distributions converged upon the first party acceleratively, which we called an avalanche phenomenon, as shown in Figs. 2 and 3. Further analysis suggested that there were two stages in decision making in this kind of situation, i. e., the subjects tended to control the outcome of the voting by voting to realize the best score cell initially in each election and shifted their choice by voting for the other party which they predicted would win at the next vote in order to realize some score.