This study aims to define the concept of Reflective Attitude in/on Conversation (RAC), based on a review of discussions in the field of reflective education, and develop a valid and reliable questionnaire to measure RAC. In the preliminary stage, three constructs, namely, Thoughtful Action (TA), Content/Process Reflection (CPR), and Premise Reflection (PR), were established, and content validity was verified by five specialists. In the exploratory stage, 467 students, who were in their 1st or 2nd year of school, were sampled from four high schools. In the confirmatory stage, 1,055 students who were in their 1st or 2nd year of school were sampled from 11 high schools across our two sessions. The results of exploratory factor analysis show that 12 among 27 preliminary items have communality greater than 0.3, factor loading to the hypothesized construct greater than 0.4, and factor loading to un-hypothesized constructs lower than 0.3. Further, the 3-factor-model’s Qui-square test does not reject the null hypothesis that there are only three factors, while the 2-factor-model’s Qui-square test rejects the null hypothesis that there are only two factors. The 3-factor-model also shows suitable TLI value (.979), while the 4-factor-model revealed a highly saturated TLI value. The results of confirmatory factor analysis with our first confirmatory stage samples (568 students), show that two items among 12 are not reliable, and NFI (.849) and CFI (.879), among five model fitness indices, do not satisfy their criteria (>.900). After revising the two unreliable items, the results of confirmatory factor analysis with our second confirmatory stage samples (487 students), reveal that all the 12 items are reliable, and although NFI (.897) still does not satisfy the criterion, CFI (.912), GFI (.936), RMR (.037), and RMSEA (.076) satisfy their criteria. In the conclusion, we suggest some implications of measured results, and some further research directions.