This paper describes a general approach for automatically tuning of reinforcement learning algorithms` parameters. In this approach a reinforcement learning agents` parameters are tuned by other more simple algorithms of reinforcement learning. We will explain this approach by tuning one of the parameters of a Q-learning and statistical clustering algorithm. The results of tuning this parameter will be described by some simple examples. Comparing the result of an algorithm using automatically tuned parameter and the algorithms with fixed parameters will show that the former is generally more flexible and capable of performing better in most cases.