The purpose of this study was to investigate a new methodology for detection of differences in middle grades students' math anxiety. A mixture partial credit model analysis was used to detect distinct latent classes based on homogeneities in response patterns. The analysis detected two latent classes. Students in Class 1 had less anxiety about "apprehension of math lessons" and "use of mathematics in daily life," and more "self-efficacy for mathematics" than students in Class 2. Students in both classes were similar in terms of "test and evaluation anxiety." Moreover, students in Class 1 were found to be more successful in mathematics, mostly like mathematics and mathematics teachers, and have better educated mothers than students in Class 2. Manifest variables of gender, attending private or public schools, and education levels of fathers did not differ among the latent classes. Characterizing differences between members of each latent class extends recent advances in measuring math anxiety.