Speech enhancement is effective in solving the problem of noisy speech. Hilbert-Huang Transform (HHT) is efficient for describing the local features of dynamic signals and is a new and powerful theory for the time-frequency analysis. According to the theory of HHT, this text introduced a new method of speech enhancement to improve the speech quantity and the signal noise ratio (SNR) of processed data. By the method of empirical mode composition (EMD), the speech signal is decomposed into several IMFs. Then remove the background noise from each IMF according to its own characters and rebuild the signal. While the SNR of the speech is low, the experiment results show that this algorithm is valid on tested noise conditions for most of speech signals and is capable to improve the SNR of the speech. Comparing with some other methods for speech enhancement such as methods based on spectrum subtraction as well as the wavelet transform, we can find that the HHT-based method is better to a certain extent.