Letter recognition is typically faster in words than ih nonwords. In t
his study, we tested the word-superiority effect obtained when either
subletter features or misalignment of letters had to be detected. Subj
ects were presented with both high- and low-frequency words and with l
egal and illegal nonwords. Space among the letters varied. In the regu
lar space condition, letters within a string were separated by normal
spaces; in the irregular space condition, letters were separated by no
rmal spaces, small spaces, or large spaces. In Experiment 1, subjects
were required to detect the presence of a bold segment contained in on
e of the letters of the string. No word-superiority effect was obtaine
d. Furthermore, spacing affected neither latencies nor response accura
cy. In Experiment 2, subjects were required to detect the presence of
a letter misaligned with respect to the others. Again, no word-superio
rity effect was obtained. However, spacing affected this task, irregul
arly spaced strings being responded to more slowly and less accurately
than regularly spaced strings. The results indicate that at the first
stages of analysis, words and non-words are similarly coded. The patt
ern obtained is consistent with a multistage model of word recognition
in which parallel feature extraction processes are followed by a leve
l at which spatial relationships are computed and by a third level at
which abstract grapheme identity is recovered from letter shapes.