A reweighted zero-attracting/repelling least-mean-squares (LMS) algorithm is proposed in this paper for sparse system identification, which attracts zero and near-zero coefficients towards zero, and repels large coefficients against zero. Specifically, an l1-norm constraint on the deviation between the unknown coefficient vector and its estimate is integrated into the original cost function of the LMS algorithm. This generates a correction term in the update equation, which always attracts the coefficients in the estimate towards their optimal values, instead of zero, as is the case in the existing zero-attracting type LMS algorithms. Simulation results demonstrate that the proposed algorithm outperforms the existing l1-norm-constraint sparsity-aware LMS algorithms in the identification of sparse systems.