An improved strategy for finding the ab initio global minimum of Si,H, clusters based on a cluster genetic algorithm (CGA) is presented. Four new mutation operators are introduced which better facilitate retention of desirable structural features between the different generations of clusters in the CGA. The new CGA generally finds global minima with fewer energy evaluations and is now able to find the problem global minima for Si14H2O and Si6H6 which were not as readily located by our previous CGA. The new CGA is used to find the global minima for the even larger Si18H24 and Si18H22 clusters. (C) 2004 Elsevier B.V. All rights reserved.