– This paper proposes a knowledge-based neurocomputing approach to extract and refine a set of linguistic rules from data. A neural network is designed along with its learning algorithm that allows simultaneous definition of the structure and the parameters of the rule base. The network can be regarded both as an adaptive rule-based system with the capability of learning fuzzy rules from data, and as a connectionist architecture provided with linguistic meaning. Experimental results on two well-known classification problems illustrate the effectiveness of the proposed approach.