Objective：To explore the forecasting method of disease index time series of cucumber downy mildew disease. Methods： Using the time series of cucumber downy mildew disease, we established an autoregressive integrated moving average model,ARIMA（2,2,0） based on model identification, comparison of residual variance, estimation and verification of parameter, observation of the correlation of the estimates matrix, autocorrelation check of the residuals, analysis of the fitting of model and so on. Results： An ARIMA model （2,2,0） was established： （1＋0. 487 1B＋0. 554 7B^2）（1-B）^2y, =α1, with the Sum of Squared Error （SSE） being 0. 001 822 and the Root of Mean Squared Error （RMSE） being 0. 008 537. The predicted values of validating date fitted well with the primary values. The established model showed satisfactory forecasting ability and was suitable for forecasting the middle stage and late stage cucumber downy mildew disease. Conclusion： Limiting the alternatives of model by residual variance, together with parameters estimation, the correlation of the estimates matrix, the autocorrelation check of the residuals and the fitting test, can help to search for suitable model quickly and accurately