The invention relates to the technical field of machine learning, discloses a method for predicting the question doing wrong position on the basis of a Bayesian network. According to method, an estimation method which can ensure the prediction accuracy of the question doing performance of an answering person or the question doing wrong position while minimize the labor cost is provided, then moreaccurate information can be provided for targeted explanation, that is to say, compared with similar methods, the estimation accuracy rate detailed to the step level can be provided, therefore, the performance of the current student on a question can be more effectively predicted, and the completion condition of the current student instead of traditional rough question-level right and wrong is evaluated; and in addition, current estimation can be corrected by collecting more facts, therefore, the collected fact number and type can be flexibly adjusted according to the needed precision, and then the method cannot be limited by application scenes basically.