In this paper we present and discuss a comparison between statistical and regional climate modeling techniques for downscaling GCM prediction. The comparison is carried out over the Capitanata region, an area of agricultural interest in south-eastern Italy, for current (1961-1990) and future (2071-2100) climate. The statistical model is based on Canonical Correlation Analysis (CCA), associated with a data pre-filtering obtained by a Principal Component Analysis (PCA), whereas the Regional Climate Model REGCM3 was used for dynamical downscaling. Downscaling techniques were applied to estimate rainfall, maximum and minimum temperatures and average number of consecutive wet and dry days. Both methods have comparable skills in estimating stations data. They show good results for spring, the most important season for agriculture. Both statistical and dynamical models well reproduce the statistical properties of precipitation, the crucial variable for the growth of crops.