Accurate prediction of precipitation plays a significant role in flood control and efficient development and utilization of water resources. Due to the strong nonlinearity and variability of precipitation series, it is difficult for the traditional statistical prediction model to accurately characterize the temporal characteristics of precipitation series.Therefore, prediction model of monthly precipitation based on complementary ensemble empirical mode decomposition with adaptive noise（CEEMDAN）and bi-directional long short-term memory（BiLSTM）was proposed in this paper. The precipitation monitoring data of Yichun meteorological station in Jiangxi province from January 1960 to December 2013 were used to establish the prediction model. The prediction results used by the CEEMDAN-BiLSTM model are compared with those of long short-term memory （LSTM）, BiLSTM, complementary ensemble empirical mode decomposition（CEEMD）-LSTM, CEEMD-BiLSTM and CEEMDAN-LSTM models. The results show that the precipitation component series with less fluctuation can be obtained based on CEEMDAN method, and the BiLSTM model constructed by this method can capture the variation characteristics of precipitation series well. In addition, the root mean square error,mean absolute error and mean absolute percentage error of the prediction results used by CEEMDAN-BiLSTM are smaller, and the correlation coefficient is larger, indicating that CEEMDAN-BiLSTM model has better performance in precipitation prediction. The model proposed by this paper can provide a new idea for precipitation prediction．