Despite increasing interest in the topic, the extent to which linguistic processing demands attentional resources remains poorly understood. We report an empirical re-examination of claims about lexical processing made on the basis of the picture-word interference task when merged in a dual-task psychological refractory period (PRP) paradigm. Two experiments were conducted in which participants were presented with a tone followed, at varying stimulus onset asynchronies (SOAs), by a picture-word stimulus. In Experiment 1, the phonological relatedness between pictures and words was manipulated. Begin-and end-related words decreased picture naming latencies relative to unrelated words. This effect was additive with SOA effects. In Experiment 2, both the semantic and the phonological relatedness between pictures and words were manipulated. Replicating Experiment 1, effects arising from the phonological manipulation were additive with SOA effects on picture naming latencies. In contrast, effects arising from the semantic manipulation were under additive with SOA effects on picture naming latencies, that is, semantic interference decreased as SOA was decreased. Such contrastive pattern suggests that semantic and phonological effects on picture naming latencies are characterized by distinguishable sources, the former prior to the PRP bottleneck and the latter at the PRP bottleneck or after. The present findings are discussed in relation to current models of language production.