Abstract: In signal time-frequency analysis, the long-time window analysis often improves the output signal-to-noise ratio and frequency resolution. However, for frequency modulation signals, the time-frequency energy concentration of the line spectrum can be reduced to affect the instantaneous frequency estimation. The generalized Warblet transform (GWT) method has a better time-frequency analysis performance than the short time Fourier transform (STFT) for frequency modulation signals, but the performance degrades or even fail in long time window analysis because of the estimation errors of the frequency modulation initial phases. To solve this problem, a time-frequency analysis method of GWT with frequency modulation initial phase compensation (FMIPC-GWT) is proposed. During the frequency modulation parameter estimation, the phase passed by half of the time window is compensated to the initial phase of the modulation components, and the accuracy of the parameter estimation is improved to make the instantaneous frequency estimation more accurate. The simulation and experimental data verify that this method has a better performance for nonlinear frequency modulation signals than the STFT and GWT methods. FMIPC-GWT method has a promising application in the line spectrum detection and instantaneous frequency estimation for frequency modulation signals.