Experience shows that there are no totally error-free complex systems. An online error detection is one possibility to avoid or at least reduce damage caused by failures. The aim is to detect errors and to apply recovery or repair before the errors can cause a failure. For this we follow a model-based approach using TSK-like fuzzy models (TSK stands for Takagi, Sugeno, and Kang, who first proposed this type of fuzzy models). In a first step, we automatically build an approximation model of the system under observation using measured input/output data (systems identification). Then, during operation, a significant deviation between the model behaviour and the measured system behaviour indicates a system error.