Source number estimation based on information theoretic criteria is ineffective in colored noise field, and the performance of the Gerschgorin radii algorithm may decline under low signal-to-noise ratio conditions. To solve these problems, a method of source number estimation using delay correlation preprocessing is proposed. The delay correlation matrix of all received data is calculated to reduce the influence of noise on the estimation performance. The eigenvectors are then grouped according to the grouping rules so that different kinds of characteristic spaces can be obtained, one containing noise eigenvectors and the other containing both noise and signal eigenvectors. Finally, the delay correlation matrix is projected onto the characteristic spaces,
and thus the source number can be estimated from the projected data. Simulation and measured data show correctness and effectiveness of the algorithm.