This study examines the sources of instability in survey measures of individuals’ racial attitudes and the differences across White, Latino, and Asian American students. Past studies on opinion stability have neglected the comparative characteristics across groups. Drawing on five-wave panel data from UCLA undergraduate students, and applied generalizability theory to decompose error variance, I find that White students tend to have larger true opinion variance, making them easier to attenuate error variance. However, for Asian American and Latino students, their opinion instability attributes to both true opinion and measurement error. I argue that a mix of partially consistent ideas and considerations in item-specificity affect Latino and Asian American students’ opinion reliability. In addition, this study also suggests that Asian American and Latino respondents need more survey items to lower the measurement error than their White counterparts.