Monte-Carlo simulations can be used as an evaluation function for Alpha-Beta in games. If w is the width of the search tree, d its depth, and g the number of simulations at each leaf, the total number of simulations is at least g × (2 × w d 2). In games where moves permute, we propose to replace this algorithm by another algorithm that only needs g × 2 d simulations for a similar number of games per leaf. The algorithm can also be applied to games where moves often but not always permute, such as Go. We detail its application to 9x9 Go.