We propose an evolutionary algorithm (EA) for global optimization of binary atomic clusters. The method is applied to binary Lennard-Jones (BLJ) clusters as a testing ground, while a limited set of results for mixed argon-krypton clusters are also reported. We show that the method is effective in discovering all global minima of BLJ clusters for nuclearities up to N = 50 and different values of the atoms size ratio. Moreover, one new global minimum has been discovered for N = 38 and size ratio 1.05. (C) 2009 Elsevier B. V. All rights reserved.