Abstract: The fundamental notion of active learning has a long history in machine learning. In this paper, an active Bayesian net model is provided and several common methods for sampling are given. Then two strategies for active learning are discussed: one is the method based on maximizing and minimizing entropy, and the other is the method combining the uncertainty sampling and minimizing classification loss. Meanwhile, also given is the method that classifies the example and update model parameters incrementally. Artificial and practical experiments show that the active learning methods proposed have high precision and recalls in a few examples with the class label.