The biggest problem in content-based image retrieval (CBIR) is a big gap between high-level semantic contents and low-level features. As an effective solution, relevance feedback has been put on many efforts for the past few years. In this paper, a new relevance feedback approach with progressive learning capability is proposed. It is based on a Bayesian classifier and treats positive and negative feedback examples with different strategies. It can utilize previous users' feedback information to improve its retrieval ability. The experimental results show that this algorithm achieves high accuracy and effectiveness on real-world image collections.