In this paper we use the cortical current density based inverse solution to classify Event Related Potential, in particular for error-related potential elicited during Brain-Computer Interface experiment. We selected discriminant cortical sources for comparing classification performance with respect to surface EEG. We found that the data from cortical sources achieves higher classification accuracy for most of the subjects. In addition, the inverse method exhibits consistently discriminant activity for the sources located over the anterior cingulate cortex region for different time points. This level of neurophysiological interpretation in terms of localisation of selected cortical sources is enabled with the use of inverse solution.