Objective: As a pilot study to develop a screening tool for eating disorders(EDs), the authors delineated factor structure and discriminant validity of the 64-item Korean version of Eating disorder inventory(EDI). Method: The EDI was administered to 1,217 Korean girls` high school students. Principal component analysis, one-way ANOVA and Receiver operating characteristic(ROC) curve analysis on EDI total and subscale scores were done. Diagnosis of EDs was based on a self report questionnaire derived from the Structured interview for DSM-IV. Result: Thirteen factors were extracted, explaining 58% of total variance but could be restructured into nine. Item construction was comparable with the Garner et al`s report except Interoceptive awareness subscale. Significant differences were noticed among diagnostic groups in the EDI total scores and subscale scores of drive for thinness, body dissatisfaction, bulimia/overeating and emotional/interpersonal problem. Other subscales such as interoceptive awareness, perfectionism, fear of maturity, interpersonal distrust and personal ineffectiveness failed to differentiated diagnostic groups. ROC curve analysis revealed dissatisfactory sensitivity and specificity of EDI as a screening tool. Moreover, EDI failed to detect AN type of EDs, showing greater usefulness for BN type. Conclusion: These data suggest that a new screening tool comprising ED- specific subscales would prove enhanced sensitivity and specificity for detecting different diagnostic groups of EDs for non-clinical population.