Eating disorders affect a disproportionate number of young females and have the highest mortality rate of all psychological disorders. To address research needs, many scales have been developed to screen for eating disorders. Among these scales, the Eating Attitudes Test (EAT) is the most utilized due to the fact that it was one of the earliest developed scales. However, a number of researchers have raised questions about its measurement properties, in particular the factor structure. The purpose of this study was to examine the construct validity of the EAT through a confirmatory factor analysis (CFA). Research participants were female undergraduates (N=785), who responded to the EAT questionnaire during the beginning of the semester. Procedures in the Windows LISREL 8.12 (Jöreskog & Sörbom, 1993) computer program were utilized for the CFA. Because the basic assumption of multivariate normality was not met, the weighted least square estimation method was applied to the model. The CFA was conducted to test two widely adopted models: three-factor model ('Dieting', 'Bulimia and Food Preoccupation', and 'Oral Control') with 26-items (Garner et al., 1982), and four-factor model ('Dieting', 'Oral Control', 'Awareness of Food Contents', and 'Food Preoccupation') with 20-items (Koslowsky et al., 1992). The goodness-of-fit indexes of the three-factor (26-item) model were within tolerable ranges (e.g. GFI=.91 and CFI=.84). However, the high root mean square error of approximation (.08) and small critical N value (155.03) signified the three-factor model was not a 'good fit' to the data. The chi-square statistic for the four-factor model was significant (c2=725.41, p=.000), and goodness-of-fit indexes of the model were acceptable (e.g. GFI=.95 and CFI=.90), indicating the model was admissible. Model determination was reached through eliminating four items with low lambda values and one factor to further improve the instrument. The final step in CFA was to test the redesigned three-factor (16-item) model through three hypothesized nested models: four-, two-, and one-factor models. Nesting procedures revealed the three-factor (16-item) model was significantly (p < .01) superior to the hypothesized two- and one-factor models. However, the hypothesized four-factor model with a new factor called 'Body Image' was significantly (p < .01) better than the three-factor model. A newly tested EAT with 16 items was thus suggested with the following four factors: 'Dieting Behaviors' (5 items), 'Awareness of Food Contents' (4 items), 'Food Preoccupation' (4 items), and 'Body Image' (3 items).Keyword(s): gender issues, measurement/evaluation, wellness/disease prevention