The purpose of this paper is to analyze individual difference and consistency in response scale use using a mixed Rasch Model(Rost, 1991), which combines latent class analysis and Rasch models. A total of 6,908 middle school students(8th graders) were included and analyses by WINMIRA(von Davier, 2001) were conducted on the academic motivation scales from KELS 2006 database. The results revealed that the two latent subgroups of respondents differing in terms of their response styles were identified in all the sub-scales. The most frequently chosen response options for class 1(mild-response style group: MLRS) are disagree or agree, whereas those for class 2(extreme-response style group: ERS) are strongly disagree or strongly agree. Second, divergent validity of the four sub-scales of academic motivation increases when response styles are taken into account. Third, the consistency of response styles across all the four sub-scales was investigated by latent class analysis. It was found that the response style was consistent for about 65% of the respondents across the four types of motivations. Practical significance of analyzing and correcting response styles in empirical data were finally discussed.