Traditional algorithms based on eigenvalue decomposition such as Akaike Information Theory(AIC) and Minimum Description Length(MDL) criterion not only require a huge amount of computation, but also reduce the performance or even cannot correctly work under the condition of low SNR or few snapshots. A novel source number estimation method based on subspace analysis is proposed. Firstly, the subspace estimation of the signal is completed fast by using Multi-stage Wiener Filter(MWF), then the projection value of array signal covariance in subspace matched filters is calculated. At last, the number of sources is obtained by analyzing their orthogonality. The simulation results show that the proposed method has better performance than other algorithms mentioned above in the low SNR or few snapshots, and the amount of calculation is greatly reduced.