Computational models that implement a serial mechanism of phonological assembly predict interactions between the size of the pseudohomophone (PsH) effect and stimulus length. Models with frequency-sensitive word representations predict baseword frequency effects. These predictions were tested in a lexical-decision task. The results showed constant PsH effects across different word lengths (in favor of parallel phonological activation) and baseword frequency effects (in favor of frequency-sensitive representations), However, the baseword frequency effect was opposite of what the models predicted. This result is most easily accommodated by models that assume an orthographic verification mechanism. The plausibility of such a mechanism was further supported by the results of 2 additional experiments investigating the effects of response speed and spelling probability (feedback consistency) on the size of the PsH effect.