Based on the data of 172 weather stations supplied by the National Meteorological Information Center and the hindcast data of the models from China, American, Japan and Europe from 1983 to 2010 and operational application results from 2011 to 2014, the prediction performance of monthly temperature is evaluated and analyzed by using Anomaly Correlation Coefficient(ACC), Trend Anomaly Inspection Evaluation (PS), and Anomaly Symbol Consistency rate (PC). The results indicate that the monthly temperature prediction performance of EC and CFSv2 is better than those of BCC and TCC model and have some skill. From space, the PC from CFSv2 is better in the first half year than in the second half year, and the range of more than 80% is larger. It is also shown that CFSv2 and EC have certain skill in typical low temperature years in summer in Northeast China