The study reports the results of a masked priming experiment with morphologically complex Russian nouns. Participants performed a lexical decision task to a visual target that differed from its prime in one consonant. Three conditions were included: (1) transparent, in which the prime was morphologically related to the target and contained the diminutive suffix -k, e.g., gorka ‘little mountain’ – gora ‘mountain’; (2) pseudo-derived, in which there was an apparent but false morphological relation between the prime and the target similar to that in the transparent condition, e.g., lunka ‘hole’ – luna ‘moon’; and (3) form, in which the phonological/orthographic overlap between the prime and the target was coincidental and could not be misanalysed as due to morphological reasons, e.g., as parta ‘desk’ – para ‘pair’. A facilitatory priming effect was found for targets in the transparent and pseudo-derived conditions but not in the form condition. The findings support the hypothesis that at an early stage of lexical processing, morphological decomposition is automatic and is not obligatorily governed by semantic transparency (Taft, 1979; Taft & Forster, 1975). Furthermore, the process of decomposition appears to apply until smallest possible morpheme-sized units are obtained.