This paper presented the results of a large-scale empirical study on optimizing the response categories of a rating scale that tapped into strategic competence in English vocabulary learning. The unique feature of the research design was that Rasch-Andrich Rating Scale Model was drawn upon in the current study to avoid the potential measurement pitfalls as existed in classical test theory. A set of diagnostic criteria were used to serve the purpose: 1) category frequency, 2) average measure, 3) category fit, and 4) step calibration. In total, six types of response formats with a variation in way of agreement were submitted for analysis. The empirical data involved 911 Taiwanese 1st year senior high school students. They were asked to fill in a well-developed instrument which targeted to measure learners' self-regulating capacity in vocabulary learning (Tseng, Dornyei, & Schmitt, 2006). The results showed that the response categories in the format of ‘disagree’ - ‘slightly disagree’ - ‘slightly agree’ - ‘agree’ functioned qualitatively better than the other five category structures. The results of the study were likely to further the understanding of the way in which ordered rating scale could be constructed in the field of SLA.