Words have been found to elicit a negative potential at the scalp peaking at approximately 400 ms that is strongly modulated by semantic context. The current study used whole-head magnetoencephalography (MEG) as male subjects read sentences ending with semantically congruous or incongruous words. Compared with congruous words, sentence-terminal incongruous words consistently evoked a large magnetic field over the left hemisphere, peaking at approximately 450 ms. Source modeling at this latency with conventional equivalent current dipoles (ECDs) placed the N400 m generator in or near the left superior temporal sulcus. A distributed solution constrained to the cortical surface suggested a sequence of differential activation, beginning in Wernicke's area at approximately 250 ms, spreading to anterior temporal sites at approximately 270 ms, to Broca's area by approximately 300 ms, to dorsolateral prefrontal cortices by approximately 320 ms, and to anterior orbital and frontopolar cortices by approximately 370 ms. Differential activity was exclusively left-sided until >370 ms, and then involved right anterior temporal and orbital cortices. At the peak of the N400 m, activation in the left hemisphere was estimated to be widespread in the anterior temporal, perisylvian, orbital, frontopolar, and dorsolateral prefrontal cortices. In the right hemisphere, the orbital, as well as, weakly, the right anterior temporal cortices were activated. Similar but weaker field patterns were evoked by intermediate words in the sentences, especially to low-frequency words occurring in early sentence positions where there is little preceding context. The locations of the N400 m sources identified with the distributed solution correspond well with those previously demonstrated with direct intracranial recordings, and suggested by functional magnetic resonance imaging (fMRI). These results help identify a distributed cortical network that supports online semantic processing.