Memetic frameworks for the hybridization of wrapper and filter feature selection methods have been proposed for classification problems. The frameworks incorporate filter methods in the traditional evolutionary algorithms to improve classification performance while accelerating the search in the identification of crucial feature subsets. Filter methods are introduced as local learning procedures in the evolutionary search to add or delete features from the chromosome which encodes the selected feature subset. Both single/multi-objective memetic frameworks are described in this chapter. Single objective memetic framework is shown to speedup the identification of optimal feature subset while at the same time maintaining good prediction accuracy. Subsequently, the multiobjective memetic framework extends the notion of optimal feature subset as the simultaneous identification of full class relevant (FCR) and partial class relevant (PCR) features in multiclass problems. Comparison study to existing state-of-the-art filter and wrapper methods, and the standard genetic algorithm highlights the efficacy of the memetic framework in facilitating a good compromise of the classification accuracy and selected feature size on binary and multi class problems.