Theories of spoken word production generally assume that mapping from conceptual representations (e.g., [furry, feline, domestic]) to phonemes (e.g., /k/, /æ/, /t/) involves both a meaning-based process and a sound-based process. A central question in this framework is how these two processes interact with one another. Two theories that occupy extreme positions on the continuum of interactivity are reviewed: a highly discrete position (e.g., Levelt, Roelofs, & Meyer, 1999), in which the two processes occur virtually independently; and a highly interactive position (e.g., Dell et al., 1997) in which the two processes exert considerable mutual influence over one another. Critical examination of the empirical data reveals that neither position can account for the full range of findings. An alternative position, the restricted interaction account (RIA), is described. By combining aspects of both highly discrete and highly interactive accounts, RIA can account for the existing empirical data, as well as for more recent challenges to interactive accounts.