Abstract: In recent years people and enterprises have paid more and more attention on the fast growth of new media blog (derived from “Web blog”) Web sites. Blogs constitute a huge blogsphere by trackbacking and recommending each other. In this blogsphere, people could freely express their opinion and feelings about topic they interested in, and could also comment on new product in the market. Retrieving blogger’s opinion on lead story and hot topic is very important for applications such as market survey, network public opinion discovery and warning. The goal of blog opinion retrieval is to retrieve the blog post that not only relate to given query but also has comment on the query. The paper introduces probabilistic inference model into blog opinion retrieval, and presents an algorithm based on probabilistic inference model. The model combines topical scoring and sentiment scoring to a uniform probabilistic inference theory model, could effectively reveal the topical facets between blog post and query and the strength of sentiments about the given query and then combine the resulting topical score and sentiment score to constitute final score. Experiment result shows that the algorithm could effectively model the topical facets and sentiments, and could also identify the opinions about given query.