Bilinear time-frequency transformation can possess a simultaneous high resolution both in the time domain and the frequency domain. It can be utilised to detect weak transient vibration signals generated by early mechanical faults in complex background and thus is of great importance to early mechanical fault diagnoses. It has been found that the spectrogram has low resolution, and there is strong cross-terms in Wigner-Ville distribution and frequency aliasing and information loss in Choi-Williams distribution (CWD). Hence, they cannot achieve satisfied transient signal detection results. To enhance the capability of detecting transient vibration signals, based on the analysis of exponent distribution, this paper presents some novel alias-free time-frequency distributions. These distributions can avoid the information loss in CWD while suppressing the cross-terms. Moreover, they have high simultaneous resolutions in both the time and frequency domain. Digital simulation and gearbox fault diagnosis experiments prove that these new distributions can effectively detect transient components from complicated mechanical vibration signals. (c) 2005 Elsevier Ltd. All rights reserved.