In the recent literature in generative morphology, it has been assumed
that multimorphemic words must be characterized as having a hierarchi
cal internal structure which is expressible in the form of morphologic
al trees. This paper reports on an experiment which addresses the foll
owing question: Are these structures relevant to complex word recognit
ion? In a naming latency experiment subjects were presented with prefi
xed and suffixed nonsense roots. The use of stimuli such as these allo
wed for the control of real word effects such as frequency and semanti
c plausibility and made it possible to systematically vary the configu
ration of the morphological trees. Significant response tune differenc
es were found between the morphologically illegal forms and legal conf
igurations. This was taken as evidence that subjects do compute morpho
logical representations. Because no differences were found between leg
al left-branching structures and legal right-branching structures, it
was concluded that morphological computation is not sensitive to the s
erial ordering of morphemes within a complex word