With a picture-picture experiment, we contrasted competitive and non-competitive models of lexical selection during language production. Participants produced novel noun-noun compounds in response to two adjacently displayed objects that were categorically related or unrelated (e.g., depicted objects: apple and cherry; naming response: "apple-cherry"). We observed semantic interference, with slower compound naming for related relative to unrelated pictures, very similar to interference effects produced by semantically related context words in picture-word-interference paradigms. This finding suggests that previous failures to observe reliable interference induced by context pictures may be due to the weakness of lexical activation and competition induced by pictures, relative to words. The production of both picture names within one integrated compound word clearly enhances lexical activation, resulting in measurable interference effects. We interpret this interference as resulting from lexical competition, because the alternative interpretation, in terms of response-exclusion from the articulatory buffer, does not apply to pictures, even when they are named.