One of the most important problems in face recognition problem is designing the feature space which represents human face the "best". Concatenating the popular feature sets together and forming a high dimensional vector introduces the curse of dimensionality problem. For this reason, feature selection is required in order to reduce the dimension of the feature space. In this study, popular feature sets used in face recognition literature are considered and comparison between these sets is done. Furthermore, high dimensional space which is obtained by concatenating all the available features is reduced to a lower dimensional space by using the Minimum Redundancy Maximum Relevance  feature selection method. ORL  and UMIST  face databases are used in experiments.