In exclusion task paradigms the participants are shown masked word stimuli with low visibility, then they must complete the stem of the target word with a novel, nonmatching word. Elevated errors (i.e., matching) suggest unconscious perception because the target is perceived, but this perceptual information is apparently not used for properly performing the exclusion task. We tested the hypothesis that exclusion failure is caused by response confusion created from highly similar responses. Response similarity was manipulated by varying the stem length with 1-letter (e.g., A), 2-letter (e.g., AC), or 3-letter (e.g., ACT) stems. Contrary to expectations, exclusion failure effects did not increase as the stem length increased from 2 to 3 letters. A higher degree of response similarity does not necessarily increase exclusion failure effects. Furthermore, the small effect from the 3-letter stem condition only weakly replicates previous studies, thereby raising doubt about the use of exclusion tasks for the study of unconscious perception.