Sentiment analysis which attempts to extract polarity from data (mostly text) has attracted many researchersin recent years. English as the richest language in this area consists of many resources and research work, but many otherlanguages including Persian lack these resources and research work. In this paper, a new hybrid approach based onclassifier combination is proposed for sentence-level and document-level sentiment analysis, which is applied on thePersian language. Moreover, language issues such as negation and intensification have been investigated. The obtainedresults approve the effectiveness of the proposed methodology.