Abstract: Two novel DOA (Direction Of Arrival) estimation methods are proposed using sparse representation when the signal number is unknown. One is the method using sparse representation based on the eigenvector of covariance matrix. The biggest eigenvector of covariance matrix is proved to be the linear combination of all steer vectors and is extracted to build sparse representation model for DOA estimation. The other is the method using sparse representation of high-order power of covariance Matrix. This method approximates the signal sub-space through the high order power of the spatial covariance matrix on the basic of signal eigenvalue being larger than noise eigenvalue. Then the column vector of high order power of the spatial covariance matrix is extracted to construct the sparse representation model for DOA estimation. The theoretical analysis and experimental results show the two methods have a better performance than the MUSIC algorithm in the aspects of accuracy, resolution and adaptability to coherent signals without estimating the number of signals.