Nowadays, the veracity, velocity, values, and size of data are growing exponentially. In fact, the data is growing beyond the capacity of current hardware facilities. This resulted in high cost of storing the data. Perhaps, some of the data stored are not very useful and create problems when mining the data, to make some sense out of it. Feature selection is a step forward towards reducing the unnecessary huge amount of the stored data. In this study, Flower Pollination Algorithm (FPA) along with its binary version (BFPA) are used for feature selection on some medical datasets. The results obtained is in favor of the BFPA with better classification accuracy of over 90% on some of the datasets and fewer number of features compared to FPA, improved harmony search with rough set together with particle swarm optimization with rough set. Hence, the experimental results demonstrate the efficiency and effectiveness of BFPA as the best technique among the evaluated methods for feature selection particularly on medical datasets.