The article summarizes results from two experimental studies on reactivity. In the first experiment, 24 college-age participants received a computerized treatment that delivered a grammar lesson, practice, and feedback on assignment of semantic functions in Latin. Verbalizations did not induce reactivity on accuracy, but they slowed down posttest performance. The second experiment (N= 24), which did not include a grammar lesson, showed that thinking aloud had a facilitating effect. Our results suggest that reactivity depends not only on the task and on the nature of the assessment tool but also on the nature of the dependent variables (i.e., latency vs. accuracy). In contrast to previous SLA studies (with the exception of Sachs and Polio, 2007), our study advises researchers to exercise caution when implementing verbal protocols in their designs.