This paper considers a simple adaptive learning rule in Bayesian games where players employ threshold strategies. Global convergence results are given for supermodular games and potentital games.