This article reviews a series of magnetoencephalographic (MEG) experiments aimed at identifying cortical areas and time windows relevant or even critical for fluent reading. The approach was to compare single-word processing in fluent and dyslexic readers. The activations which differed between the two groups were then studied in more detail to determine their functional roles. In fluent reading, overall visual feature processing occurs about 100 milliseconds (ms) after seeing a word, in the posteromedial extrastriate cortex bilaterally. This activation does not differentiate between letters and symbols. The first reading-specific signal is detected about 150 ms after word onset, when the left inferior occipitotemporal cortex responds preferentially to letter strings. After 200 ms, the left superior temporal cortex, in particular, is engaged in semantic processing of single words and their integration with connected text. While visual feature processing seems to be within normal limits in dyslexic subjects, reading is disrupted during the first 200 ms after seeing a word, at the letter-string specific stage. The subsequent activations are weak and delayed as compared with those in fluent readers. Also presented is a case of deep dyslexia, where the same tools were used to demonstrate that reading comprehension was still subserved by the left hemisphere despite severe damage.