There is little agreement on the mechanisms involved in second language (L2) processing of regular and irregular inflectional morphology and on the exact role of age, amount, and type of exposure to L2 resulting in differences in L2 input and use. The article contributes to the ongoing debates by reporting the results of two experiments on Russian verb generation and recognition in a lexical decision task (LDT) with priming by highly proficient late L2 learners, early interrupted (heritage) learners, and adult native speakers of Russian. Inflected verbs varying in regularity, type, and token (lemma) frequency were used. The experiments document the role of obligatory decomposition and complex allomorphy involved in (de)composition and mapping as well as type and token frequency in L1 and L2 verb generation and lexical access, with no sharp division between regular and irregular verbs. The results are inconsistent with either the dual-system or the single-system approach to morphological processing, and are compatible with "hybrid" theories combining rule-based decomposition and input-frequency-based probabilistic mechanisms. All of the verb types showed priming effects, both in native and nonnative lexical access. However, the degree of facilitation depended on decomposition costs for different verb types only in nonnative participants. The study also shows differences in early (heritage) and late L2 learners. In the production task, heritage speakers outperformed L2 learners in real, but not novel, verb generation and in the use of the regular default pattern, whereas L2 learners showed an advantage in the use of the cue-based complex morphological pattern. In the priming task, heritage speakers were faster than late L2 learners and insensitive to morphological complexity, whereas L2 learners showed longer latencies in response to the verbs with complex inflectional morphology. These differences are associated with the differences in their respective language ling backgrounds. It is tentatively suggested that these two groups of learners may rely on different processing mechanisms and, possibly, neural paths. (Contains 2 figures, 5 tables and 28 notes.)