We investigated the early moments of visual word recognition,when the retinal information—by hypothesis split verticallyalong the fovea—is divided into two visual pathways,projecting the right visual field into the left hemisphere (LH),and the left visual field into the right hemisphere (RH).Wearing red/blue anaglyph glasses, participants performed alexical decision task to compounds (FOOTBALL) andmonomorphemic words that were either pseudo-compounds(CARPET) or unsegmentable (JINGLE). The stimuli werepresented (masked, 60 ms exposure) in three colorcombinations: all black, red/blue (ipsilateral visual pathways),and blue/red (contralateral pathways). For the red/blue andblue/red conditions, the colors were split either at themorpheme boundary (legal split) or at a character to the left orto the right of the split (illegal split). We found an advantage(RT and accuracy) of compounds over non-compounds,independent of pathway, and an advantage of legal vs. illegalconstituent split. Results suggest that the visual wordrecognition system performs parsing analyses that are inconsonant with the morphological objects of the language. Theadvantage of pseudo-compounds over unsegmentablessuggests that at an early—pre-lexical—stage the system ispartially insensitive to the semantic properties of the wholeword.