A new method for speech enhancement based on Kullback-Leibler (K-L) divergence has been presented in this paper. First, the algorithm performs wavelet-packet transform to noisy speech and decomposes it into sub-bands; then we apply a threshold on coefficients in each sub-band to obtain enhanced speech. To determine the threshold, first the distributions of noisy speech, noise and clean speech coefficients are calculated; then a symmetric K-L divergence between the noisy speech and noise distributions is calculated. Finally a speech/noise decision is made based on the calculated distance. We conducted some tests using TI.MIT database in order to assess the performance of the proposed method and to compare it to previous speech enhancement methods. The algorithm is evaluated using the Perceptual Evaluation of Speech Quality measure (PESQ) and the output SNR. We obtain an improvement of up to 2.2dB on SNR and 1.2 on PESQ for the proposed method in comparison to the results of the previous wavelet based methods.