Six experiments that were designed to test the adequacy of criterion bias explanations of the word frequency effect and the semantic priming effect are reported. It was found that criterion bias models correctly predicted higher error rates in a lexical decision task for nonwords that were misspelled versions of high-frequency words (e.g., MOHTER), rather than low-frequency words (e.g., BOHTER). Also correct was the prediction of increased error rates for misspelled words preceded by a semantically related word (e.g., NURSE-DOTCOR). However, in a misspelling decision task (in which the subject must decide whether the stimulus is a word, a misspelled word, or a nonword), it can be argued that criterion bias should be inoperative, since correct responses must be delayed until all orthographic information has been checked; this should eliminate both frequency and semantic priming effects. This was found not to be the case; clear frequency and priming effects were obtained for both words and misspelled words.