Backward Induction Method, which is the most basic algorithm for game tree search, has two weak points. The move selected by Backward Induction Method is assured the best move as far as the search depth of the game tree but not necessary the best move at the end of the game. Also the evaluation values for leaf nodes do not necessarily present the best advantage at the end of the game.?In this paper for a game tree search in endgame phase we propose a new algorithm, probability method, which is useful on the game over with the constant moves such as Othello game. We show that probability method is more effective than Backward Induction Method for the selection of just before two players can search till the end of the game.