Recent information appliances becomes very sophisticated, while their operations are not so easy for a general user. In the present situation, a user is forced to adapt to a machine, and the users must enter correct key-input sequences for target operations. To solve the problem it is desired to develop a user support system where a machine estimates a human intention of a target operation and the user's operation ability, and supports adaptively according to the estimate results. In this paper, a building method of a Bayesian Network (BN) for intention estimation of an information appliance user is proposed. BN is effectively employed for inference based on learning using practical data. Causal relation between an intention, operation ability, and a feature of incorrect key-input subsequences are required to be reflected for building BN, and we must establish appropriate nodes for its expression. Especially, it is important to express the causal relation between an intention and correct/incorrect key-input subsequences.