BACKGROUND: Nonlinear mutual cross prediction (MCP) characterizes dynamic interdependence among nonlinear systems. MCP also reveal relative strength of the coupling between systems, thus provides information about the direc-tion of interdependence. The aim of this study is to apply MCP algorithm to multi-channel EEG and to characterize spatio- temporal pattern of seizure. METHODS: We analyzed MCP of EEG of three medically intractable temporal lobe epilepsy patients, who underwent temporal lobectomy (left 2, right 1). Asymmetry of nonlinear cross predictability between channels was investigated. Five epochs of interictal EEG free from epileptiform discharge(s) and of ictal EEG were analyzed. RESULTS: In interictal period, both frontal and occipital region appeared a weak driving force while awake and this driving force was further weakened during sleep. Before the onset of the seizure (preictal phase), the intensity of driving system became slightly stronger around seizure foci in 3 out of 8 seizures while no significant change was seen on the naked eyes. However this change was dim and not continuous. At the onset of seizure (ictal phase), 5 out of 8 seizures showed strong driving force around seizure foci. Three seizures without significant change initially had strong driving force as synchronous seizure discharges became built-up and spreading to surrounding areas in the middle of seizure. All seizures showed ipsilateral frontotemporal strong driving force and centroparietal response system, which was typical spatio-temporal distribution of MCP. CONCLUSION: MCP analysis may be a useful method for detecting spatio-temporal distribution and propagation pattern in temporal lobe epilepsy.