We read words one at a time. Crowding has shown that words are recognized by parts. Are these parts also recognized one at a time? When you have more information (context) about what the next word might be, you need fewer parts to identify it. We find that the time required to recognize the next word is roughly proportional to the log of the number of possibilities. Results with RSVP presentation of letters, digits, and words reopen the question of serial letter recognition that was thought to be closed one hundred years ago.