Fault features in a mechanical system typically appear as transient impulses which can be analyzed using time-frequency analysis (TFA) methods. Recently, a new TFA technique termed as time-reassigned multisynchrosqueezing transform (TMSST) was proposed to capture the transient impulses in a signal for fault diagnosis. However, the TMSST was developed based on the short-time Fourier transform (STFT) which could yield unclear high-frequency image due to the fixed sliding time window used in STFT. To overcome this limitation, TMSST is combined with the S-transform and local maximum method in this study for an improved time frequency representation in the signal analysis. An extractive reconstruction algorithm binding the maximum value of the spectral envelope is further proposed for spectral decomposition. A simulated noise-added signal and four experimental bearing defect data are used in the study to verify the validity and effectiveness of the technique developed in a step-by-step manner. The results confirm that the proposed technique can accurately extract the fault features for bearing operated under constant or varying speed conditions.