Adopting Linacre's guidelines for evaluating rating scale effectiveness, we examined whether and how a six-point rating scale functioned differently across raters, speech acts, and second language (L2) proficiency levels. We developed a 12-item Computerized Oral Discourse Completion Task (CODCT) for assessing the production of requests, refusals, and compliment responses among 109 examinees of L2 Chinese. Their oral productions were evaluated by two L1 Chinese raters based on a holistic rating scale simultaneously tapping communicative function, situational appropriateness, and grammaticality. Rating scale functioning differed across raters, speech acts, and proficiency levels. Such variations were caused by multiple factors: (1) the two raters interpreted the rating scale differently, (2) the generic rating scale was unable to represent the nuances in different speech acts, (3) the two proficiency groups drew on different portions of the rating scale, and (4) the rating scale categories were redundant due to an excessive focus on the descriptive rather than the interpretive function during rating scale development.