To solve the real-time through-the-wall imaging problem in the presence of wall ambiguities, an approach based on the least-squares support vector machines (LS-SVMs) is proposed. This technique converts the through-the-wall problem into the establishment and use of a mapping between the backscattered data and the target properties. The wall parameters and the propagation effects caused by the walls are both included in the mapping and can be regressed after the LS-SVM training process. The target properties are estimated using LS-SVM. Noiseless and noisy measurements are performed to demonstrate that the approach can provide comparable performance in terms of robustness and efficacy, as well as improved performance in terms of accuracy and convenience, in comparison with the approach based on the support vector machine (SVM). The influence of the radius of the target on the estimation problem is discussed, and the estimated results show that both the LS-SVM and the SVM have good performances in terms of generalization.