This thesis examines the degree to which lexical and nonlexical procedures for word naming represent distinct processing strategies. A series of experiments were aimed at testing the hypothesis that grapheme-phoneme conversion is relatively more attention demanding than lexical processing (Paap & Noel, 1991). Contrary to predictions of a slowing of the nonlexical route, word naming in Experiment 1 was not affected by a concurrent digit memory load. Experiment 2 failed to support the prediction that the lexical route is susceptible to interference from a visual dot pattern load. In Experiment 3, standard word naming without a memory load produced similar effects of word frequency and regularity to those found in the memory load conditions. The failure to observe shifts in processing strategy in response to load manipulations is tentatively attributed to the predominance of lexical processing due to the nature of disyllabic words. In Experiment 4, a digit load failed to modulate consistency effects but naming latencies decreased with increasing load, as did nonword naming latencies in Experiment 5. It is suggested that readers strategically lower the criterion for initiating a pronunciation in response to task difficulty. Finally, phonological decision latencies in Experiment 6 slowed down from low to high load when at least one item was a pseudohomophone or a nonword. The results imply that nonlexical processing is attention demanding when an accurate phonological code must be assembled in the absence of lexical information. The lack of concurrence costs on word latencies suggests that a relatively automatic lexical procedure may predominate in generating word-specific phonology. Contrasting effects of load are interpreted as indicating distinct lexical and nonlexical strategies and are taken to support dual-route models of word recognition.