This paper presents and discusses an approach of user modeling. A set of neural networks is utilized to store, maintain and infer users' task-related characteristics. Such networks function as associative memories that can capture the causal relationships among users' characteristics for the system adaptation. It is suggested that this approach can be expected to overcome some inherent problems of the conventional stereotyping approaches in terms of pattern recognition and classification. It can also avoid the complexity of truth maintenance in default reasoning that is required in previously known stereotyping approaches.