Brain-Computer Interface (BCI) technology provides a new mode of direct communication between man and machine, replacing human's normal output control pathways of nerves and muscles. It embeds huge potential to be explored in medicine, rehabilitation and entertainment that can be used for the healthy as well as the disabled. In order to thoroughly exploit BCI's capabilities, efficient man-machine interaction has to be established by accurately identifying human intentions from brain waves. Electroencephalogram (EEG) is a strong tool in BCI research as it is a cheap and easy non-invasive recording methodology of brain activity encoded with human thoughts. This paper investigates an EEG dataset recorded from 85 healthy subjects performing imagination of motor movements, and highlights the necessity of selecting the subject-specific spatial and spectral features to optimize motor imagery recognition performance. Subject-specific spectral and spatial features associated with right and left hand motor imagery are identified on account of the discriminative weights of EEG signals recorded from motor cortex region. Discriminative capability has been estimated using the Fisher ratio values of each frequency component for each channel. Using the proposed hybrid subject-specific selection of channels and bands, the proposed BCI system is capable of offering comparable classification accuracy with the state of the art methodology which employs more number of channels and frequency bands.