In this work we investigate the application of the wavelet analysis to improve the well-known horizontal to vertical spectral ratio (HVSR) technique for site effect estimation. Wavelet analysis has been widely used in the last decade for its capacity to localise the signal in both the time and frequency domains. The wavelet packet transform and the evaluation of a cost function for every node permit one to remove from the signal the least important components and lead to significant improvements in the identification of the fundamental frequency. In this paper a new algorithm is presented based on wavelet analysis, that allows, in some cases, to improve the identification of the peak in the QTS ratio. A number of examples of application are also presented.