One of the challenges in EEG-correlated fMRI studies is the selection of a representative time-course from the EEG data, which can be used to derive a predictor of the BOLD signal recorded in each voxel using fMRI. Independent Component Analysis (ICA) is commonly used for this purpose, but the selection of meaningful components is mostly performed through visual inspection by a neurophysiologist, which renders it both time-consuming and subjective. A new methodology is presented here for the automatic selection of independent components from the EEG, called Spatio-Temporal Templates for Independent Component Selection (STTICS). Dataset-specific temporal and spatial templates for the epileptic activity are first extracted. The correlation between these templates and the components’ time-courses and corresponding topographies, respectively, is then computed and used as input to a k-means clustering algorithm. The components belonging to the cluster with the highest correlation coefficients are classified as epilepsy-related. The performance of the proposed method, STTICS, was compared with the one of the only other existing method in the literature for the same purpose (COMPASS), in terms of accuracy (ACC), true positive rate (TPR) and true negative rate (TNR), by simulations with artificial data and application to real data. A total of 19 real datasets acquired from 6 patients with drug-refractory focal epilepsy undergoing pre-surgical evaluation were studied, and the neurophysiologist selection was considered as the ground truth in this case. In general, STTICS outperformed COMPASS in both simulations and real data. In the epilepsy datasets, the components automatically selected by STTICS were mostly consistent with the selection by the neurophysiologist, with a performance of ACC=96.3%, TPR=65.9% and TNR=98.5%. The ability of our method to accurately and objectively select epilepsy-related independent components makes it an important contribution in simultaneous EEG-fMRI epilepsy studies.