This paper proposes a new statistical approach based on frequency singular value decomposition (SVD) to enhance the SNR of the noisy multicomponent seismic wavefield. Our filtering algorithm consists of three main steps: Firstly, the frequency transformed multicomponent seismic wavefield data is rearranged into one long vector containing information on all frequencies and all component interactions. Secondly, the reduced dimensional spectral covariance matrix of the long vector data is estimated by means of singular value decomposition. Finally, the separation of the primary seismic waves from the noise is achieved by projecting the dominant eigenvector that has the highest eigenvalue of the reduced dimensional covariance matrix onto the long data vector. The experimental results have shown that the proposed algorithm outperforms the conventional separation technique in terms of accuracy and complexity.