Abstract: In this work, linear and nonlinear downscaling are developed to establish empirical relationships between the synoptic scale circulation and observed rainfall over the southeastern Brazil. The methodology used involved the outputs from ETA regional model, where prognostic equations for local forecasting were developed using artificial neural networks (ANN) and multiple linear regression (MLR). The final objective is the application of such prognostic equations to Eta output to generate forecasts rainfall. In the first experiment the predictors were obtained from ETA model and the predictand was the rainfall data from meteorological stations in southeastern Brazil. In the second experiment observed rainfall on the day prior to the forecast was included as a predictor. The definition of the predictors variables was based on synoptic analysis over the studied region and were selected the vertical component of the wind in 850 hPa, air temperature at 1000 hPa, relative vorticity in 250 hPa, precipitable water, sea level pressure, and moisture convergence at 850 hPa. The Threat Score (TS) and BIAS used to quantify the performance of the forecasts through statistical downscaling, showing that ANN was superior to MLR in most of the seasons and periods of the year. When compared to ETA forecasting it was observed that during the summer ANN has a tendency to forecast moderate and high rainfall with greater precision. Also when the observed rainfall of the prior day to the forecast is included as a predictor, the TS performed better in the forecast, above all in continuous rain and well organized meteorological systems. On the other hand, in the winter period, characterized by slight rain, ANN showed better ability in forecasting than the ETA model. The results obtained also suggest that in winter rainfall is more predictable because convection is less frequent, and when this occurs the forcing is dynamic instead of thermodynamic.