The early diagnosis of abrupt and tendentious faults in industrial robots are in the scope of the paper. A complete design environment is outlined, which provides to choose fault detection technique, and contains isolation methods and a simulation environment. This enables the users to compare the performance of various fault detection methods, and to evaluate the performance of the diagnosis system in case of different faults. The design environment is built around a systematic design methodology, which includes the setting up and the development of the dynamic model of the robot manipulator, the generation of the stochastic system model and the identification of its noise inputs, the evaluation and the improvement of the filter convergence under normal operation. The performance of the designed diagnosis system can be verified by simulating mechanism and sensor faults.