Abstract: This work performes forecast of daily rain for the ZCAS-VCAN summer pattern over 12 places in the Sao Paulo state using a statistical downscaling of the Eta/CPTEC model. An artificial neural network is used to build two models for scaling transfer. For the first case, patterns of ZCAS-VCAN were selected to generate the forecast model, and for the second one, it was considered all series of the summer period. The result of the forecast using the downscaling has shown that the rain forecast is better when the weather pattern is specified. The errors with respect to ETA model are reduced until in 60%. This feature is expected, since the dowscaling process is applied to improve the model forecasting.