Many behavioral models of the comprehension of suffixed words assume a dual-route mechanism in which these words are accessed sometimes from the mental lexicon as whole units and sometimes in terms of their component morphemes (such as happi+ness). In related neuropsychological work, Ullman et al. (1997) proposed a dual-route model for past tense processing, in which the lexicon (used for access to irregularly inflected forms) corresponds to declarative memory and a medial temporal/ parietal circuit, and the rule system (used for computation of regularly inflected forms) corresponds to procedural memory and a frontal (including Broca's area)/basal ganglia circuit. We used functional MRI and a memory encoding task to test this model for derivationally suffixed words, comparing those words that show evidence of decompositional processing in behavioral studies (-ness, -less, and -able words) with derived words that do not show decomposition effects (-ity and -ation words). By examining Broca's area and the basal ganglia as regions of interest, we found that "decomposable" derived and inflected words showed increases in activity relative to nondecomposable suffixed words. Results support a dual-route model of lexical access of complex words that is consistent with the Ullman et al. proposal.