Abstract: Sentiment analysis method can mine valuable information from a mass of automotive reviews, which has great application value in automotive product design and brand marketing. For the requirements of fine-grained analysis, a fine-grained sentiment analysis algorithm is put forward based on the entity. Firstly, the automotive reviews are preprocessed, then the model of Linear-chain CRF is used to do sentiment entity recognition and sentiment classification. Secondly, in order to relate the entity recognition with sentiment classification, the model of Linear-chain CRF is improved, and a method of two-level CRF proposed. Experimental results show that two-level CRF is better than Linear-chain CRF in sentiment analysis, which can meet the demand of fine-grained sentiment analysis of automotive reviews.