Least-squares reverse-time migration (LSRTM) is a recently developed imaging algorithm, where the image is produced using an iterative inversion process. Tests on synthetic and real data have shown the promise that LSRTM can improve the image quality by balancing the reflector amplitudes, suppressing migration artifacts and enhancing the image resolution. However, each iteration of the process is comparable in computational cost to a conventional RTM and typically 10 to 20 iterations are required to converge; thus the application of LSRTM has been limited. In this paper we incorporate statistical sampling with LSRTM to reduce its computational cost. The empirical results suggest that this approach reduces the cost of LSRTM to 2 or 3 times that of conventional RTM while retaining most of the quality improvements.