Internet is the major information source for investors. The opinion leaders' opinion can be widely spread on the Internet among investors and alter their altitude and behavior. This paper uses social network analysis technique to identify opinion leader in the online stock forum and the text analysis technique to classify the type of opinion leaders' posted message. Based on the sample data from 2011 to 2017, it is found that opinion leader sentiment can predict individual stock return in the next 1 month and up to 6 months. The reason of return predictability is not because opinion leader knows the value-relevant information on listed firms. Further studies show that the investors' identification on opinion leader is irrelevant with their prediction accuracy. Investors place more attention on the opinion leaders with positive opinion and ignore their correctness. In addition, investors repel opinion leaders who make correct negative predictions. Our study reveals the impact of opinion leaders on the market and their formation mechanism, and also provide theoretical basis for task of investor education and protection.