A previous study Valdes-Sosa and Bobes, 1990) described a negative ERP
component evoked by mis-matches in a face-feature matching task, usin
g photographs of real faces. This component could be N400 (or an analo
gue), elicited by associative priming within a non-linguistic domain:
that of face structure. To confirm this it is necessary to demonstrate
that semantic/linguistic recoding was not a necessary condition in tr
iggering the negativity. This means falsifying what we call the 'primi
ng by proxy' hypothesis, and locating the triggering mis-match within
face structure. In this paper subjects studied artificial schematic fa
ces over several sessions, and 1 week later were presented with a face
-feature matching task with simultaneous ERP recording. Since no seman
tic information or verbal labels were available, eliciting a mis-match
negativity with these faces contradicts the 'priming by proxy' hypoth
esis. In a first experiment, in which the subjects learning was contro
lled through a face familiarity decision task, no significant mis-matc
h negativity was found. However, in a second experiment in which learn
ing was controlled through a forced-choice face-feature match, a signi
ficant mis-match negativity was found in the subsequent recording sess
ion. This result supports the idea that a component similar to N400 ca
n be elicited by an associative mis-match restricted to the face-struc
tural domain.