Purpose: We aimed to build a classification system that uses resting-state (no visible scalp epileptic activity) EEG-based directed functional connectivity values to assign a patient to one of three classes: left TLE (LTLE), right TLE (RTLE) or healthy control.\r\n\r\nMethods: Twenty LTLE, 20 RTLE and 35 healthy controls underwent resting-state high-density EEG. For each subject, sixty 1-sec epochs free of artifacts or interictal spikes were selected. The source activity was obtained for 82 regions of interest using an individual head model and distributed linear inverse solution. The summed outflow and whole-brain directed functional connectivity were estimated in the theta, alpha and beta frequency bands using Granger-causal modeling. A Random Forest classifier (an ensemble of decision tree classifiers) was then used to assign the subject to one of three classes. The mean classification accuracy was computed with a leave-one-out procedure. We selected a maximum of six connectivity values for classification, using a greedy forward selection algorithm. Finally, three classifiers were built: ‘Control vs. LTLE’, ‘Control vs. RTLE’ and ‘LTLE vs. RTLE’. In the final classification system, a new subject is assigned to the class that was most voted by these three classifiers.\r\n\r\nResults: The ‘Control vs. RTLE’ classifier achieved an accuracy of 78.2% (sensitivity: 80.0%, specificity 77.2%), ‘Control vs. LTLE’ an accuracy of 83.6% (sensitivity 85.0%, specificity 82.9%) and ‘LTLE vs. RTLE’ an accuracy of 85.0% (sensitivity 85.0%, specificity 85.0%). Combining these classifiers into one system yielded that 16, 15 and 27 subjects were correctly classified as being, respectively, RTLE, LTLE and control.\r\n\r\nConclusion: The high accuracy achieved demonstrates the potential of resting-state EEG-based directed functional connectivity for the diagnosis and lateralization of TLE. This could constitute a new clinical biomarker for surgical candidates and earlier in the course of the disease.\r\n