The main objective of this dissertation is to study wind generation forecasting models using data from five wind farms, more accurately compare the performance of linear and nonlinear models. Using the methodology of the nonlinear model STAR-TREE (Smooth Transition Autoregression Tree) and compare with the linear model BoxandJenkins through statistical measures. Basically the model STAR-TREE is a combination of models STAR (Smooth Transition Autoregression) and CART (Classification and Regression Tree), thus creating a modeling tree where the transition between regimes is done smoothly through the logistics function and in the terminal nodes are adjusted predictive models. In this study will fit in the terminal nodes, a simple model of constant and a autoregressive models.