Abstract: In this paper we show that for discounted MDPs with discount factor the asymptotic rate of convergence of Q-learning is if 1/2 and otherwise.