In this paper, we propose the use of techniques computationally intensive linked to statistical methods Bayesian aimed at statistical modeling of sample data distributions, and distributions of extreme values, with the main objective the evaluation of uncertainties associated with climate change projections precipitation and air temperature in South America, by focusing on specific areas (Amazon, Brazilian Northeast, central Brazil and the River Plate Basin), within the scope of CLARIS-LPB Project. Whereas the climate is modulated by decennial patterns of sea surface temperature (SST), the objective is to determine the future climatology taking into account this fact condition and its relation to the influence due to anthropogenic factors. For these investigations, seven regional climate models outputs will be used (RCA, PROMES, MM5, RegCM3, REMO, LMDZ and Eta), associated with some sets of observational data and products derived from observations. Because these regional models are distinguished from each other, their rounds show a positive characteristic that is to group different types of configurations and parameterization, which helps expand the field of sampling possibilities related to future climate conditions (unknown factor). Thus, as the central point of the research, Bayesian scenarios are scanned future climate uncertainties that distinguish possible positive and negative phases of TSM as conditioning data for climate change, since all simulations of CLARIS-LPB models were generated with the same scenario A1B (balanced) of greenhouse gas emissions for the twenty-first century. Thus, we propose the application of simple Monte Carlo method and Markov Chain (MCMC) through the sampler algorithm Gibbs.