The paper addresses a problem of system modelling in a fuzzy environment. Firstly a general methodological framework is proposed and in the sequel it is specialized in form of an iterative procedure comprising three basic steps, namely structure determination, parameter estimation and model validation. A state-of-the-art of fuzzy modelling is discussed in detail: some advantages and existing drawbacks are carefully analyzed. Also a variety of forms of fuzzy models are handled in a uniform fashion by means of fuzzy relational equations. In the sequel attention is focused to the determination of a fuzzy relation of the model on the basis of input-output pairs of data available. The paper summarizes existing approaches useful for solving a set of fuzzy relational equations being of analytical as well as numerical character. Finally an original concept of fuzzy confidence interval expressing relevance of the fuzzy model constructed on the basis of collected data is introduced. In one extreme case (perfect fit of the data to the model) this confidence interval reduces to a single value of membership function, while in the second extreme situation the confidence interval expands into a whole \([0,1]\) interval (which indicates complete lack of model relevance).