An approach to detect, localise and classify various types of power quality disturbances in noisy conditions is proposed. The approach is based on the statistical properties of white noise. The first step is to extract a given disturbance from the original signal including the white noise. After that, a short-time correlation algorithm is applied to limit the effect of white noise on wavelet analysis. In order to enhance the detection outcome, squared transform coefficients of the analysed power signal are used. Results related to common disturbances in electrical power quality analysis have shown the effectiveness and the feasibility of the proposed approach.