Differential electrophysiological effects for regular and irregular linguistic forms have been used to support the theory that grammatical rules are encoded using a dedicated cognitive mechanism. The alternative hypothesis is that language systematicities are encoded probabilistically in a way that does not categorically distinguish rule-like and irregular forms. In the present study, this matter was investigated more closely by focusing specifically on whether the regular–irregular distinction in English past tenses is categorical or graded. We compared the ERP priming effects of regulars (baked–bake), vowel-change irregulars (sang–sing), and “suffixed” irregulars that display a partial regularity (suffixed irregular verbs, e.g., slept–sleep), as well as forms that are related strictly along formal or semantic dimensions. Participants performed a visual lexical decision task with either visual (Experiment 1) or auditory prime (Experiment 2). Stronger N400 priming effects were observed for regular than vowel-change irregular verbs, whereas suffixed irregulars tended to group with regular verbs. Subsequent analyses decomposed early versus late-going N400 priming, and suggested that differences among forms can be attributed to the orthographic similarity of prime and target. Effects of morphological relatedness were observed in the later-going time period, however, we failed to observe true regular–irregular dissociations in either experiment. The results indicate that morphological effects emerge from the interaction of orthographic, phonological, and semantic overlap between words.