This paper proposes a mechanism of noise tolerance for reinforcement learning algorithms. An adaptive agent that employs reinforcement learning algorithms may receive and accumulate many rewards for its actions. However, the amount of rewards received by the agent is not a guarantee of convergence to an optimal policy of action due to the noises produced by the environment. Therefore, we propose a noise tolerance mechanism which is able to estimate convergent policies without causing delays or an unexpected speedup in the agent?s learning. Experimental results have shown that the proposed mechanism is able to speed up the convergence of the agent achieving good action policies very fast even in dynamic and noisy environments.