With the boost of online reviews，sentiment polarity classification rises in response to the requirement of retrieving consumers＇ positive or negative opinions on certain products． The primary goal of this research is to improve the accuracy of sentiment polarity classification at the level of paragraphs for Chinese online reviews． With a view to the ways of expression and the grain of corpus，this paper presents a method to predict the sentiment polarity of Chinese online reviews in paragraphs based on sentence level sentiment analysis． Firstly，traditional classification methods are applied to predict the sentiment polarity of sentence． Then， three different algorithms i．e．，the equal weight，correlation degree and assumption of sentiment condition， are employed to calculate the contribution that each sentence lying in the different positions of paragraph makes to the sentiment polarity of paragraph． Finally an experiment has been made based on hotel and mobile phone online reviews with lengths beyond two sentences． The result shows that the accuracy of sentiment polarity classification at the level of paragraph is remarkably increased by the method proposed in this paper，by taking correlation degree of expression and assumption of sentiment condition into consideration．