We evaluated the accuracy and interobserver variability of selected ictal and postictal behavioral changes. Three observers, blinded to clinical history, EEG, and side of surgical resection, analyzed videotapes of 166 seizures in 38 patients, looking for lateralizing signs. Twenty-seven patients with temporal lobe resections were seizure-free for > or = 1 year postoperatively, and 11 with extratemporal resections had at least 90% reduction in seizures > or = 1 year postsurgery. The epileptogenic region (ER) was lateralized by analyzing lateralizing signs in 78% of patients; positive predictive value (PPV) was 94% (90% CI = 87% to 100%). Overall kappa was 0.68. Signs were considered present if seen by two or more observers. Forty-five percent had version, ie, forced and sustained head deviation (kappa = 0.76, PPV = 94%); 37% had dystonic posturing of the upper extremity (kappa = 0.47, PPV = 93%); and 34% had unilateral mouth deviation (kappa = 0.83, PPV = 92%). These signs indicated a contralateral ER. Twenty-one percent had unilateral upper extremity automatisms, all ipsilateral to the ER (kappa = 0.65, PPV = 100%); 21% had postictal dysnomia, indicating a dominant-hemisphere ER (kappa = 0.89, PPV = 100%); and 16% had ictal speech, usually indicating a nondominant-hemisphere ER (kappa = 0.75, PPV = 83%). Dystonic posturing, postictal dysnomia, ictal speech, and unilateral upper extremity automatisms may indicate a higher probability of temporal lobe epilepsy. Analysis of lateralizing signs shows good interobserver agreement and provides useful clinical information.