The system of on-line condition monitoring and fault diagnosis has put forward strict requirement in real-time storage and remote transmission. As a common method in data processing, wavelet analysis has been widely used in data compression and achieved good effect. MMethods of signal compression are presented based on wavelet firstly. According to analysis on the characters of the vibration-signal of rotating machinery and analysis on several key attributes of wavelet basis, a new selection method of optimum wavelet basis in vibration-signal compression of rotating machinery is proposed. The compression effect of corresponding wavelet basis is compared by using a practical fault vibration signal. The results show that, the signal-to-noise ratio of reconstructing signal can be improved effectively by choosing an appropriate wavelet basis function. The biorthogonal wavelet basis with lower vanishing moment in decomposition and higher regularity reconstruction is better fitting for complex vibration-signal compression of rotating machinery