In theoretical models of self-regulated learning, calibration is one important component in successful learning. Two issues of calibration are explored. First, Nelson (1987) suggested the "G" (gamma) coefficient is the most appropriate measure of calibration (judgment accuracy) and rejected signal detection theory's "d'" statistic because data commonly challenge distributional assumptions. We empirically examined this issue, comparing "G" and "d'." Second, we examined whether a learner's calibration varies across three domains of knowledge: general, word, and mathematics. A sample of 266 university students volunteered to participate in the study. Participants were selected from various undergraduate and graduate courses. Participants first answered demographic items. Then they completed three knowledge tests (general, word, and mathematics) and judged correctness for each answer provided. Order of domains was randomly counterbalanced among participants. Results show that "d'" is a valid measure of calibration, that assumptions about underlying distributions can be tested, and that preliminary evidence suggests that "d'" may be a superior measure of accuracy compared to "G." Finally, calibration varied by domain.