Face recognition in JPEG compressed domain is one of the recent challenges in biometric systems, leading to a considerable reduction in computational overhead caused by decompression process, without any notable degradation in the recognition rates. In this paper, the potential of using a limited number of lowest frequency coefficients in JPEG compressed domain face recognition is investigated, to obtain a considerable reduction in computational complexity, as well as improving the recognition rates. The proposed method preselects lowest frequency coefficients including all of the DC coefficient and a limited number of the AC coefficients for face recognition process. Simulations have been done on four of the FERET databases using different number of preselected coefficients. Experimental results show that the proposed method outperforms existing methods in recognition rate, as well as in computational and space complexity aspects.