Now online reviews have become an important resource for mining users'opinion and refining products. As a foundation of further analysis, features extraction influences the precision of the opinion mining results. This paper proposes a PMI-Bootstrapping algorithm which realizes extracting product features from Chinese online reviews by combining three language rules. First, utilize the language rules to get a candidate feature set. Then, calculate the weighted average PMI for each candidate feature with the seeds in the initial seed set. Add the candidate feature which satisfies the threshold to the seed set. Iterate until the seed set is convergent. Output the seed set as the extraction result. Experimental results show that the algorithm achieved very good precision and recall rate.