A modified harmony search(MHS) algorithm is proposed for solving large-scale system reliability problem. This algorithm amends the searching mechanism of HS algorithm, which takes the best-so-far solution as a study subject, randomly selects different dimensions to conduct improvisation, and modifies the adjustment method of parameter bandwidth(BW) to balance global and local searching. The classical large-scale system reliability problem is solved. Numerical results show that the proposed MHS algorithm is better than all the reported 6 kinds of HS algorithms. The MHS algorithm has better optimization performance on the whole compared to some excellent algorithms reported for solving large-scale system reliability problems in the recent year.