Influenced by factors such as different discharge mechanisms and transmitting routes， partial discharge signals can have multiple waveforms. Under such circumstances， single scaling function based wavelet cannot work. To solve the problem， in this paper a multiple scaling functions based multi-wavelet algorithm is introduced. A large number of simulations have been carried on in the paper for different forms of partial discharges. Noise-reduction performance of the proposed multi-wavelet algorithm under different PD modes is investigated by using the mean square error as the criterion to achieve optimal parameters in noise reduction. Extracting of multi-mode partial discharge signals and comparison with previously published results are also presented. The results obtained both in simulation and tests demonstrate that the multi-wavelet based algorithm with proper parameters can extract multi-mode partial discharge signals effectively， significantly improve the performance of existing optimal wavelet based and wavelet-set based algorithms.