In order to model uncertainty involved into a real-world process, a general approach combining both analytical and expert knowledge-based techniques to identify a multi-input multi-output non-linear system model is proposed. The approach leads to using a new type of stochastic dependence of random processes, involving corresponding expert knowledge, and which is an extension of the dispersion identification technique based on a new – expert knowledge-based – measure of stochastic dependence. The paper is preceded with a state-of-the art analysis of available knowledge-based identification and control approaches.