The derivation of local scale information from integrations of coarse-resolution General Circulation Models (GCM) is generally referred to as {\it downscaling}. The numerous downscaling approaches existing today may be classified into two broad groups, statistical and dynamical. In this paper the most relevant statistical downscaling techniques are described and some concrete examples are worked out in detail. The aim is not to present an exhaustive review of the applications of the different statistical techniques, but rather to give the potential user of experiments with GCMs an overview of the limitations of these statistical techniques for the estimation of regional climates and to suggest the essentially different options that have been developed to overcome these limitations. The existing statistical downscaling techniques are classified into three groups: linear methods, classification methods and deterministic non-linear methods. Their performance in a particular example, winter rainfall in the Iberian peninsula, is compared to a simple downscaling analog method. It is found that the analog method performs in general equally well than the more complicated methods, and it can be applied to both normal and non-normal distributed local variables. On the other hand linear multivariate methods offer a clearer physical interpretation that supports more strongly its validity in an altered climate. Downscaling analysis can be also used to validate regional performance of climate models. In addition to focusing on the correct reproduction of means and standard deviations of regional climates, the realistic relationships between the simulated large-scale climate and the regional climate, i.e. the covariability between large-scale and regional-scale, is underlined.