ERPs were recorded from subjects performing semantic and rhyme matchin
g tasks using either spoken words, printed words or pictures as stimul
i. Mismatches enhanced N400 (in the semantic task) and N450 (in the rh
yme task). Onset and peak latencies were shorter for N450 than for N40
0 with spoken words; this relationship was inverted for pictures. Thus
these latencies could index availability of semantic and phonological
codes. For printed words, the latencies were shorter for N400 than N4
50, a result that supports direct-access models of reading with late p
honological code activation. The longer latencies found for N400 and N
450 to pictures could suggest longer initial decoding for pictures wit
h respect to words.