Within a discrete two-stage model of lexicalization, semantic errors and errors of omission are assumed to be independent events. In contrast, cascading and interactive models allow for an influence of word form on lexical selection and thus for an inherent relationship in accounting for both error types. A group of 17 aphasic patients was assessed with a naming test controlling for semantic competition of the target items. Semantic errors were more frequent for targets with many competitors than for targets with few competitors while omissions were more frequent when few competitors were available. However, the overall sums of errors in both item groups were comparable. These results imply a common source of both error types and thus speak against a strictly serial model of naming.