Abstarct: This thesis investigates the fault tolerant control from different aspects, by using the Reinforcement Learning methods. First, the robustness of the Reinfircement Learning is studied. It is demonstrated that the represented approach has the robustness feature. Then, through identifying the fault-free (normal) system by using the Radial Basis Function (RBF), the residual is calculated. The main innovation of this thesis is combining the Sliding Mode control concept and the Eligibility Traces algorithm, one of the most applicable algorithms in the Reinforcement Learning frxamework, for the fault tolerant control. The proposed structure is compared with the previous one, a combination of the Artificial Neural Networks and the Sliding Mode control. It is shown that the performance of the proposed structure is superior than the previous one. The second innovation of the thesis is using the Time Window to calculate the amplitude and determining the type of the fault. The resulting data and states of the system are used to implement the fault tolerant control by using the Eligibility Traces algorithm. Finally, it is revealed that the proposed structure has a better performance in addition to the simplicity of the tolerant control. The basic methods of the Temporal Difference learning, including the Q-learning algorithm and SARSA are compared to demonstrate the fault tolerant control behavior of the proposed structure. In The last innovation, the Feedforward neural network is used to identify the normal and faulty systems, baxsed on the residual calculation. baxsed on the calculated residual, continuous Reinforcement Learning in combination with the Auto-Step algorithm is employed for the fault tolerant control purpose. Proposed methods are applied to diabetic and melanoma patient systems. There is no need of the environment model in RL and it is one of the most important features of the Reinforcement Learning algorithm in the system control. Since there is no access to the real model (patient), the mathematical models of diabetic and melanoma patients are utilized. To treat the melanoma in the presence of the fault, the half-life of the drug is considered, as well. This is an important consideration for problems which contribute with determining the optimal dosage and studied in a few researches.