© 2015, Universidad de Valparaiso. All rights reserved. Into different areas (e.g., academic, public) predictions of climate as part of the process of decision-making are required. Despite such information need, the inconsistency of global models to predict the state of the climate in small scales (regions) is widely recognized. Considering this, we tested predictions of sea surface temperature (SST) in 10 marine regions off the coast of Mexico. Using classification and regression trees, Mexican coastal states were grouped accordingly to their similarity in instrumental records of air temperature (AST). Such AST groups were considered explanatory variables together with regional climatic scale indices (e.g., Pacific Decadal Oscillation, PDO). Historical patterns of change (period, amplitude and phase) of AST and climate indices were characterized, and then its relationship with SST was analyzed using generalized additive models (GAM). The SST response to climatic scenarios was evaluated with 3 different forcing criteria. The GAM models showed significant fits and relatively high values of R2 and deviance. Projections of regional climate variability showed substantial differences in comparison to the monotonic increase in SST global models outputs. The rescaling strategy applied in this work for Mexican seas surface temperature, proved to be useful to integrate the historical variation with different forcing criteria.