We have developed a new algorithm, FINE, to enhance the spatial resolution and localization accuracy for closely-spaced sources, in the framework of the subspace source localization. Computer simulations were conducted in the present study to evaluate the performance of FINE, as compared with classic subspace source localization algorithms, i.e. MUSIC and RAP-MUSIC, in a realistic geometry head model by means of boundary element method (BEM). The results show that FINE could distinguish superficial simulated sources, with distance as low as 8.5 mm and deep simulated sources, with distance as low as 16.3 mm. Our results also show that the accuracy of source orientation estimates from FINE is better than MUSIC and RAP-MUSIC for closely-spaced sources. Motor potentials, obtained during finger movements in a human subject, were analyzed using FINE. The detailed neural activity distribution within the contralateral premotor areas and supplemental motor areas (SMA) is revealed by FINE as compared with MUSIC. The present study suggests that FINE has excellent spatial resolution in imaging neural sources.