I examined the nature of morphological decomposition in a series of masked-priming experiments with Russian prefixed nouns. In Experiments 1A and 1B, I tested 3 types of prime-target pairs in which the prime was a morphologically simple word, and a facilitation was found when the prime and the target were truly morphologically related (e.g., "narost" [outgrowth] and "rost" [growth] are morphologically related via the prefix "na"- [on]) or apparently morphologically related (e.g., "priton" [den] and "ton" [tone] seem to be morphologically related via the prefix "pri"- [toward], but this relation is false) but not when the relation was purely orthographic (e.g., "kumir" [idol] and "mir" [peace]; "ku"- is not an existing prefix of Russian). These results suggest that all orthographic forms that can be exhaustively parsed into a prefix and a stem are decomposed into (apparent) constituent morphemes during their retrieval from the lexicon. This early segmentation process is driven by morpho-orthographic but not by morphosemantic considerations and applies even for derived forms that are more frequent than their stem. (Contains 11 footnotes and 9 tables.)