In this paper, a data-driven methodology for the development of virtual models of a wind turbine is presented. To demonstrate the proposed methodology, two parameters of the wind turbine have been selected for modeling, namely, power output and rotor speed. A virtual model for each of the two parameters is developed and tested with data collected at a wind farm. Both models consider controllable and noncontrollable parameters of the wind turbine, as well as the delay effect of wind speed and other parameters. To mitigate data bias of each virtual model and ensure its robustness, a training set is assembled from ten randomly selected turbines. The performance of a virtual model is largely determined by the input parameters selected and the data mining algorithms used to extract the model. Several data mining algorithms for parameter selection and model extraction are analyzed. The research presented in the paper is illustrated with computational results.