This paper presents an improved SVD-based method for wall clutter mitigation in through-the-wall radar imaging. The dominant wall singular components are identified from the singular value spectrum. A subspace projection method is then applied to remove the strong wall clutter, residing in the dominant singular components, and separate the target signal from noise. The remaining wall clutter residual, which is mixed with the target signal, is suppressed by segmenting the range profile of the signal residing in the subspace orthogonal to the wall and noise subspaces. A Gaussian mixture is used to model the range profile, and the optimum segmentation threshold is found by minimizing the Bayes error. Experiments results show that the proposed method is more effective at reducing wall clutter and preserving the targets than some of the existing wall clutter mitigation methods.