Three experiments were run to test a model in which location is encode
d better for letter strokes than for letters. A string of four letters
or four single-stroke symbols resembling long letter strokes appeared
on each trial, with the target indicated by a cue in the postexposure
mask. The four were chosen from a set of 12 possible items of each ty
pe, with letters and symbols never mixed within a display. An error co
uld involve either the intrusion of an item not in the array, or the m
islocation of one appearing at other than the target location. In all
three experiments the predominance of mislocations over intrusions was
greater for letters, with the reverse true for symbols, supporting be
tter location coding for letter strokes than for whole letters.