The conjugate decomposition (CD), which was given for symmetric and positive definite matrices implicitly based on the conjugate gradient method, is generalized to every m × n matrix. The conjugate decomposition keeps some SVD properties, but loses uniqueness and part of orthogonal projection property. From the computational point of view, the conjugate decomposition is much cheaper than the SVD. To illustrate the feasibility of the CD, some application examples are given. Finally, the application of the conjugate decomposition in frequency estimate is given with comparison of the SVD and FFT. The numerical results are promising.