The foreign learners are not easy to learn Chinese as a second language. Because there are many special rules different from other languages in Chinese. When the people learn Chinese as a foreign language usually make some grammatical errors, such as missing, redundant, selection and disorder. In this paper, we proposed the conditional random fields (CRFs) to detect the grammatical errors. The features based on statistical word and part-ofspeech (POS) pattern were adopted here. The relationships between words by part-of-speech are helpful for Chinese grammatical error detection. Finally, we according to CRF determined which error types in sentences. According to the observation of experimental results, the performance of the proposed model is acceptable in precision and recall rates.