With the advance of the internet, e-commerce systems have become extremely important and convenient to human being. More and more products are sold on the Web, and more and more people are purchasing products online. As a result, an increasing number of customers post product reviews at merchant websites and express their opinions and experiences in any network space such as internet forums, discussion groups, and blogs. So there is a large amount of data records related to products on the Web, which are useful for both manufacturers and customers. Mining product reviews becomes a hot research topic, and existing researches mostly base on product features to analyze the opinions. So mining product features is the number one step to further reviews processing. In this thesis, we present how to mine product features efficiently and accurately. The proposed extraction approach is different from the previous methods because we only mine the features of the product from opinion sentences in which the customers have expressed their positive or negative sentiment. In order to find opinion sentences, a SentiWordNet-based algorithm is proposed. There are three steps to perform our task: (1) Identifying opinion sentences in each review which is positive or negative via SentiWordNet; (2) Mining product features that have been commented on by customers from opinion sentences; (3) Pruning feature to remove these incorrect features. Compared to previous work, our experimental result achieves higher precision and recall. It executes fast enough for practical use.