Radar imaging using multiple input multiple output systems are becoming popular recently. These applications typically contain a sparse scene and the imaging system is challenged by the requirement of high quality real-time image reconstruction from under-sampled measurements via compressive sensing. In this paper, we deal with obtaining sparse solution to near- field radar imaging problems by developing efficient sparse reconstruction, which avoid storing and using large-scale sensing matrices. We demonstrate that the "fast multipole method" can be employed within sparse reconstruction algorithms to efficiently compute the sensing operator and its adjoint (backward) operator, hence improving the computation speed and memory usage, especially for large-scale 3-D imaging problems. For several near-field imaging scenarios including point scatterers and 2-D/3-D extended targets, the performances of sparse reconstruction algorithms are numerically tested in comparison with a classical solver. Furthermore, effectiveness of the fast multipole method and efficient reconstruction are illustrated in terms of memory requirement and processing time.