ree experiments investigated the e ects of semantic and sensory information on the relationships between recognition accuracy and con dence ratings by comparing the calibration curves consisting of subjective rates of old and new responses with those consisting of nominal rates of old and new responses. In these experiments, college students n 273 intentionally studied words, then received a recognition test and rated con dence in their recognition responses. To investigate the e ects of semantic information on the calibration curves, the similarities between old and new items were varied. Experiments 1 visual and 2 auditory used the same presentation modality of study and test words. Experiment 3 used an auditory presentation of study words and visual presentation of test words. Analyses of the nominal responses showed poor or negative calibration curves by the semantic confusing as preceding results. In contrast, the analyses of the subjective responses revealed good calibration curves, if only either of semantic or sensory information could be available for recognition judgments. e present results indicate the importance of subjective analysis of recognition responses.