The pentad mean temperature anomaly forecasting model of the Xinjiang area is established using Empirical Orthogonal Function decomposition (EOF), Ensemble Empirical Mode Decomposition (EEMD), and Least Square Support Vector Machine (LSSVM) methods. Using the EEMD to decompose the first three time series obtained from EOF into a series of Intrinsic Mode Function (IMF), the predicted IMF is acquired by using LSSVM. The acquisition of pentad mean temperature anomaly is from the forecast of time series and temporal reconstruction. Tendays mean temperature anomaly is acquired based on the pentad mean temperature anomaly, and the month mean temperature anomaly was acquired based on the 10days mean temperature anomaly. The model using the temporal anomaly correlation coefficients (ACC), the skill score (SS) and anomaly sign score (〖WTBX〗R〖WTBZ〗) are evaluated. The results show that the model worked well in the first 20 pentads with an average ACC of 0.32, average SS of 0.70, and average〖WTBX〗 R 〖WTBZ〗of 050. It worked well in the first ten 10day periods, with an average ACC of 0.50, average SS of 0.50, and average 〖WTBX〗R〖WTBZ〗 of 0.50. It also worked well in three months, with an average ACC of 0.50, an average SS of 0.50, and an average 〖WTBX〗R〖WTBZ〗 of 0.80.