The majority of ensemble creation algorithms use the full set of available features for its task. Feature selection for ensemble creation has not been carried out except for some work on random feature selection. In this paper we focus our attention on genetic based feature selection for ensemble creation. Our approach uses a genetic algorithm to search over the entire feature space. Subsets of features are used as input for ensemble creation algorithms. In this paper we compare boosting and bagging techniques for ensemble construction together with feature selection approaches. Also we compared the memory employed for the ensembles using the well-known C4.5 induction algorithm for ensemble construction. Our approach show more reliable ensembles with less than 50% of the total number of features employed.