The NCEP Regional Spectral Model (RSM), with horizontal resolution of 60km, was nested in ECHAM4.5 AGCM simulations forced with observed SSTs over Northeast Brazil. An ensemble of 10 runs for the period of January-June, 1971 - 2000 was done. The NCEP RSM is able to retain the large-scale information produced by the ECHAM4.5 AGCM and partially corrects an AGCM bias in displacement of the Atlantic ITCZ. Observed local scale precipitation patterns and their interannual variability are captured by the RSM as well. Good potential prediction skills are demonstrated by the RSM in predicting localized precipitation variability. The RSM probabilistic distribution function of precipitation anomalies at the seasonal time scale is in better agreement with the observations than that of the ECHAM AGCM. The RSM is also capable to resolve the observed precipitation intraseasonal variability, such as precipitation intensity distribution and dry spells. A drought index and a flooding index were defined to indicate the severity of drought and flooding conditions, and their interannual variability was well reproduced by the RSM. The overall RSM performance in the downscaled climate of ECHAM4.5 AGCM is satisfactory over Nordeste. The primary deficiency is that the RSM has a systematic dry bias for precipitation simulation, which can be largely corrected by statistical tools.