An information-theoretic approach for studying synchronization phenomena in experimental bivariate time series is presented. "Coarse-grained" information rates are introduced and their ability to indicate generalized synchronization as well as to establish a "direction of information flow" between coupled systems, i.e., to discern the driving from the driven (response) system, is demonstrated using numerically generated time series from unidirectionally coupled chaotic systems. The method introduced is then applied in a case study of electroencephalogram recordings of an epileptic patient. Synchronization events leading to seizures have been found on two levels of organization of brain tissues and "directions of information flow" among brain areas have been identified. This allows localization of the primary epileptogenic areas, also confirmed by magnetic resonance imaging and pasitron emission tomography scans.