Abstract: The stability of leading Empirical Orthogonal Function (EOF) modes in summer rainfall is evaluated by using the cross-validation and some other methods on the basis of 160 station-observed monthly precipitation values recorded in China during 1980-2012, and the potential advantage of EOF utilized in the seasonal forecasts of China summer rainfall is discussed. Our results suggest that when a one-year sample is randomly removed, the first four EOF modes suggest higher stability. The EOF-based potential forecast skill is mainly in South China, and the 33-year-averaged anomaly correlation coefficient (ACC) between the idealized forecast and the observed field is approximately 0.6. In contrast, the stability of the leading EOF modes for the precipitation anomaly percentage is greatly affected by extreme climate events. However, when we artificially removed this effect, the stability of the first three EOF modes improved, where the potential forecast skills appear in most parts of China, and the averaged ACC between the potential forecast and observed rainfall anomaly percentage is 0.48. The stability of the EOF mode exhibits an obvious decrease when the forecast period is extended, particularly for the last three EOF modes, which suggests that forecasting skill scores would decrease when the EOF method is applied to predict the future rainfall anomalies two years in advance.