An automated process for selecting sites for new wind farm installations is proposed. The region of interest is divided into a 1-km-square mesh, and geographical data such as altitude and wind speed are used to sort the mesh cells into regions that are feasible for wind farm installations. Before grouping the meshes, feasible meshes for constructing wind farms are extracted using a set of constraints. We tested two different constraints for grouping the feasible areas, either by maximizing the annual mean wind speed or by minimizing the covariance between the power outputs of each cell in the group. The first strategy is more attractive if the goal is to meet an expected level of power output each year, while the second strategy is intended to supply the most-stable power. Portfolio theory was then applied to the evaluate efficient-frontier curves of the two site-selection results from the mean and variance of the total expected power outputs. The analysis showed that grouping unit areas to maximize average wind speed most effectively suppresses variance in the expected output of an installation, and efficiently distributes the optimum wind farm locations.