The power curve modeling for wind turbines is a key tool used to predict the generated electric power, and to monitor and operate wind turbines, directly affecting the investments in wind farms. However, the collected data can sometimes contain outliers, which can significantly affect the modeling. A classic technique widely used for this purpose is called Polynomial Regression and recently, the use of the Multilayer Perceptron Network (MLP) has been gaining prominence. More recently, the Extreme Learning Machine (ELM) was developed, and it has been validated as a plausible solution for global modeling. In this paper, in addition to the mentioned techniques, the following Artificial Neural Networks (ANN) are also presented: the Self-Organizing Map with K winners (KSOM) and the Local Linear Mapping (LLM), which are based on local modeling. These models are implemented using Ordinary Least Squares (OLS) and M-estimator techniques to obtain the parameters used in the estimation and to verify the effect of the outliers on the input data. By employing qualitative and quantitative analysis, the obtained results prove that the proposed models based on local modeling are a realistic alternative to the conventional methods for estimating the power curve in wind turbines.