Several studies have proposed the use of inverse solutions based features to improve the decoding performance of brain-computer interfaces. Most of these studies have compared the performance of inverse solutions features over scalp activity in a small set of electrodes. However, the estimated sources are indeed a linear combination of scalp-wide activity. Therefore, this comparison may be biased against surface EEG. Performance comparison in three ERP-based protocols show that classifiers combining larger sets of EEG electrodes may perform comparably, and previous reports may have overestimated the advantages of using inverse solution based features.