Objective : The objective of this study was to test the reliability of functional magnetic resonance imaging (fMRI) evaluation of memory function in clinical practice to predict postoperative memory decline in patients with refractory medial temporal lobe epilepsy (MTLE) candidate to surgery. Methods: Twenty-six consecutive patients with MTLE who underwent a complete presurgical evaluation were included. All patients underwent fMRI memory study and complete neuropsychological assessment. Lesions consisted in hippocampal sclerosis in 18 patients (12 right and 6 left), dysembryoplastic neuroepithelial tumor (DNET) in 5 cases (4 right, 1 left), epidermoid cyst in one patient (right). Two patients had no lesion (2 left). Results: Nineteen patients (73%) underwent surgery. The other seven patients (27%) declined surgery, mainly because of the risk of memory deficit. The fMRI procedure correctly predicted both verbal and nonverbal memory postoperative outcome in 13 of the patients (72%), failed to predict a postoperative memory worsening in only two patients (12%), and predicted worsening in three patients (17%) that remained stable (versus 44%, 39%, and 17% with the sole neuropsychological testing). The reliability of the fMRI procedure was not influenced by the type of lesion, the side of the epileptic focus, or the type of preoperative memory profile (typical or atypical). Significance: Appearing as a valuable clinical tool to predict postoperative memory outcome, fMRI may add information over and above other available tests.