Abstract: The paper aims to evaluate the cross-season forecasting ability of seasonal climate model and multimodel ensembles on precipitation during flood season in Heilongjiang, and improve the accuracy rate of climate prediction. Based on the data of 3 seasonal climate models in China, the United States and Europe from 1983 to 2017, by applying the multi- model ensembles to the prediction of precipitation during flood season in Heilongjiang, 4 quantitation methods including anomaly correlation coefficient (ACC), comprehensive testing of trend anomalies (Ps), graded score (Pg) and anomalous symbol agreement rate (Pc), we comprehensively evaluated the ability of the above 3 seasonal climate models and multi-model ensembles to predict crossseason precipitation during flood season, and found an objective forecast method suitable for the precipitation during flood season in Heilongjiang. The results showed that: each model had a certain cross-season forecasting ability to forecast the precipitation in Heilongjiang during flood season, but had a relatively poor forecast ability for the anomalous magnitude of precipitation trend. According to the comparison of the prediction scores of each model, EC model is relatively better and could be applied in the forecasting, multi-model super ensemble is better than the daily business quality score and the multi-model equal weighted average ensemble, which could be used as a reference for climate prediction during flood season.