Hilbert-Huang transform (HHT) is an excellent time-frequency analysis method, which contains empirical mode decomposition (EMD) and Hilbert transform (HT). A given signal is decomposed by EMD and then each of the obtained intrinsic mode functions (IMFs) is demodulated by HT. Aim to solve the problems of envelope errors, mode mixing and end effort of EMD method and negative frequency of HT, an improved HHT method (IHHT) is proposed, which contains generalized EMD (GEMD) and improved direct quadrature (IDQ) demodulation. By GEMD several different IMFs are obtained by defining different mean curves and in each rank the best IMF is selected. The improved empirical AM-FM decomposition and IDQ are used to demodulate the component. The proposed method improved the accuracy of decomposition and demodulation and restrained the end effort. IHHT is applied to rolling bearing experiment data and the analysis results indicate that IHHT is an effective signal processing method.