Neurosurgical interventions often require the presurgical determination of language dominance or mapping of language areas. Results obtained by fMRI are closely correlated with invasive procedures such as electrical stimulation mapping or the intracarotid amobarbital test. However, language fMRI is not used routinely, because postprocessing is time-consuming. We utilized a real-time analysis software installed directly on the MR console computer and SPM99 as reference postprocessing software. We assessed the reliability of the immediate determination of language dominance based on individual activation maps by comparing the results of the visual analysis of images derived from conventional postprocessing with those produced by the real-time tool. All images were rated independently by six senior neurologists blinded to other data. We validated the robustness of the real-time method statistically by comparing global and regional lateralization indices derived from real-time and postprocessing analysis. Functional MRI was performed with a standard 1.5-T whole-body scanner. Brain activity was contrasted between an alternating semantic judgment and letter matching task. Twelve right-handed, healthy control subjects and 12 consecutive patients with drug-resistant, localization-related epilepsy were investigated. The semantic condition induced almost invariably left hemispheric activations in Broca's area, the premotor cortex, the dorsolateral prefrontal cortex, and the temporoparietal region. Although real-time analysis reduced noise less effectively than SPM99, visual ratings and lateralization indices produced highly concordant results with both methods. In conclusion, real-time fMRI, as used here, allowed reliable language lateralization and mapping in less than 15 min during routine clinical MRI investigation with no need for postprocessing.