For an objective interpretation of cerebral metabolic pattern to find epileptogenic zones in patients with temporal lobe epilepsy (TLE), the authors developed a computer-aided classifier using an artificial neural network (ANN). They studied 261 epilepsy patients diagnosed as no abnormal findings (NA, n=64), left TLE (n=116), or right TLE (n=81) on interictal brain F-18-fluorodeoxyglucose positron emission tomography (FDG PET) by the consensus of two expert physicians. Seventeen asymmetry indexes between the mean counts of the 34 mirrored regions were extracted from the spatially normalized images and used as input parameters. The three diagnoses of NA, left TLE, and right TLE were used as outputs of the ANN. The structure of the ANN was optimized with variable error goals and the number of hidden units. On the criteria of agreement of diagnoses with those of expert viewers, the best classifier was chosen, which yielded a maximum average agreement of 85% for the test set when the authors used an error goal of 20 (sum of squared error) and ten hidden units. The authors could devise an ANN that performed as well in diagnosing left or right TLE on FDG PET as human experts and could be used as a clinical decision support tool.