The view that impairments of attention may constitute an important factor underlying impaired language performance in aphasia has gained support in recent years. Aphasiologists taking this view have generally proceeded from resource allocation models of attention, with little or no attention given to alternative models. One alternative model of dual-task performance is the central bottleneck model, which proposes a single-channel limit at response selection or other central processing stages. The first purpose of the present experiments was to further examine the effects on word production of lexical frequency in the context of the psychological refractory period (PRP) dual-task method. The second purpose was to examine whether the reaction time (RT) patterns obtained under conditions promoting equal task emphasis are more consistent with the central bottleneck or central resource models. Three dual-task experiments were conducted using speeded picture naming and tone identification tasks presented at varying timulus onset asynchronies (SOAs). In experiment 1, lexical frequency affected primary-task naming and secondary-task tone identification RTs approximately equally. In experiment 2, lexical frequency affected secondary-task naming RTs similarly at all levels of SOA, after potentially confounding variables were taken into account. It was concluded that frequency-sensitive lexical processing in picture-naming participates in the central processing stage of the dual-task models under study. In the third experiment, the two tasks were presented in variable order and subjects were instructed to give equal attention to both. On tone-primary trials, tone RTs increased with decreasing SOA, a result consistent with the central resource model and inconsistent with the central bottleneck model, unless augmented by the assumption that participants grouped responses on short SOA trials. Also, additional analyses restricted to those participants demonstrating a lexical frequency effect on the secondary naming task found that lexical frequency and SOA interacted on primary-task tone RTs such that tone responses preceding low-frequency naming responses were slower than those preceding high-frequency names. This further suggests that these subjects allocated more central processing capacity to the naming task on low-frequency trials. Comparison of results across the three experiments suggested that participants in Experiment 3 demonstrated less dual-task interference than predicted by either model.