Efforts to harness wind energy on a large scale have gained momentum across the world. By the end of December 2013, a cumulative capacity of more than 300 GW of wind power projects had been installed all over the world. One of the key aspects involved in implementing wind power projects is the analysis of wind speeds distributions observed or recorded and assessment of annual energy output from the wind turbines. The wind speed frequency distribution is generally assumed to follow two-parameter Weibull Distribution. In general, across the world, annual energy generation estimations of a wind turbine at a given site are assessed on the basis of Weibull Distribution. However, in this paper, based on a robust analysis carried out on over 208 measurement sites in India, we show that multi-peak Gaussian distribution functions are a significantly improved representation of observed wind speed distributions.