The topography of auditory event-related potentials (ERPs) was examined during 3 kinds of tasks: selection of a specified real word or nonsense syllable from a list; simple detection of each of the same stimuli without discrimination; and classification of a set of words according to a specified semantic category. The potentials that were associated with the additional processing required by the discriminative tasks were disclosed by subtracting the wave forms obtained in the detection condition from those obtained during discriminative performance. Difference wave forms were also derived between the semantic classification and verbal discriminative ERP to delineate the changes associated with the extraction of word meaning. The topography of the ERP associated with stimulus detection was comparable to that found in previous studies of evoked potentials to non-speech stimuli. This distribution was consistent with 2 cortical generators, one within the supratemporal plane and the other on the lateral surface of the superior temporal gyrus. When discriminative performance was required on the basis of acoustic stimulus properties, the topography of the difference wave form that reflected this discriminative processing extended more posteriorly over temporal cortex. Semantic processing elicited a further posterior extension of ERP components by 330 msec after stimulus onset, as well as longer latency potentials that were not present in the verbal selection task. These differences imply that a more extensive portion of language cortex is engaged in semantic classification than in verbal identification.