We propose a new framework for conceptualizing measures of instructed second language (L2) pronunciation performance according to three sets of parameters: (a) the constructs (focused on global vs. specific aspects of pronunciation), (b) the scoring method (human raters vs. acoustic analyses), and (c) the type of knowledge elicited (controlled vs. spontaneous). Adopting this model (Framework for L2 Pronunciation Measurement) as a synthetic tool, we coded the instruments found in 77 studies of L2 pronunciation teaching published between 1982 and 2017. We calculated the frequency of each measurement type and reexamined the interaction of instructional effectiveness and measurement within the sample. According to the results, instruction is most effective when it targets learners' monitored production of specific segmental or suprasegmental features. The efficacy of instruction remains relatively unclear when gains are measured globally via subjective, human judgments, especially at a spontaneous level.