In this paper, an improved two-stage simulated annealing algorithm is presented for the minimum linear arrangement problem for graphs. This algorithm integrates several distinguished features including an efficient heuristic to generate good quality initial solutions, a highly discriminating evaluation function, a special neighborhood function and an effective cooling schedule. The algorithm is evaluated on a set of 30 well-known benchmark instances of the literature and compared with several state-of-the-art algorithms, showing improvements of 17 previous best results.