The study was designed to examine the construct validity and internal consistency reliability of the Eating Attitudes Test (EAT) using a confirmatory factor analysis (CFA). Two widely adopted EAT models were tested: three-factor (Dieting, Bulimia and Food Preoccupation, and Oral Control) with 26 items (Garner, Olmsted, Bohr, & Garfinkel, 1982), and four-factor (Dieting, Oral Control, Awareness of Food Contents, and Food Preoccupation) with 20 items (Koslowsky et al., 1992). Research participants included two samples of female college students (calibration N = 785, cross-validation N = 298). Maximum Likelihood estimation method was adopted. The fit indexes from the three-factor EAT-26 represented unacceptable model fit (RMSEA = 0.11, SRMR = 0.11, CFI = 0.73, AGFI = 0.74). Similarly, the fit indexes from the four-factor EAT-20 model provided a poor fit (RMSEA = 0.09, SRMR = 0.07, CFI = 0.85, AGFI = 0.83); however, after eliminating four items with low factor loadings, the four-factor EAT model with 16 items was found to have an acceptable fit (RMSEA = 0.08, SRMR = 0.05, CFI = 0.91, AGFI = 0.88). The EAT-16 model was then crossvalidated on an independent sample and was found to have acceptable configural and metric invariance as well as internal consistency reliability. (Contains 2 figures and 7 tables.)