EAT-26: 26 questions about food, weight and control
The EAT-26 asks how often a set of thoughts and behaviours around food have applied to you. Garner and colleagues built it in 1982 by cutting the 40-item original down to the 26 items that loaded most strongly in a sample of 160 women with anorexia nervosa and 140 female comparison subjects. The short form correlates .98 with the long one, so very little was lost in the trimming.
Answers are frequency-based and scored unevenly: always counts 3, usually 2, often 1, and sometimes, rarely and never all count 0. Item 26 is scored the other way round. The total runs 0 to 78, and the whole thing takes about eight minutes.
Three subscales
| Subscale | Items | What it covers |
|---|---|---|
| Dieting | 13 | Avoiding high-fat food and pursuing a thinner body |
| Bulimia and food preoccupation | 6 | Preoccupation with food and loss of control over eating |
| Oral control | 7 | Self-control around eating and pressure from others to gain weight |
The three factors together explain 40.2% of the total variance. Dieting and bulimia sit at opposite poles: restriction shows up across anorexia, while binge episodes appear in a subset of people. The factor structure is not stable across populations, and non-clinical samples have produced three, four, five and even seven factors. A Rasch analysis found the items work best in the middle to high range and separate low-risk people poorly.
The cutoff, and why it moves
A total of 20 or more is the original threshold, taken from discriminant analysis that classified 83.6% of cases correctly. The longer EAT-40 managed 84.9% with its own cutoff of 30, so the short form gave up almost nothing. Later samples pushed the line around: a Japanese study put the optimal cutoff at 17 (AUC 0.925, sensitivity 0.866, specificity 0.868), and work in overweight and obese groups landed on 11.
The scale is a screening step, not a diagnosis. Garner's own protocol pairs the total with five behavioural questions and BMI before deciding whether a referral makes sense, which is a useful reminder that the questionnaire alone is not the whole picture.
Reliability
Alpha was 0.90 in the anorexia sample, and the dieting factor on its own also reached 0.90. A meta-analysis of 14 studies put the pooled alpha at 0.85 (95% CI 0.81 to 0.88), and found language, age and gender made no difference to the estimate. A Japanese sample reported 0.90 in the clinical group (95% CI 0.87 to 0.92) and 0.83 in the non-clinical group (0.79 to 0.86).
Chinese version
Wang and colleagues (2015) revised the scale with 801 and 420 university students and ended up with 19 items and four factors instead of 26 and three: dieting, bingeing and food preoccupation, awareness of food content, and compensatory behaviour. Model fit was good (chi-square/df 2.93, GFI 0.91, CFI 0.90, NFI 0.85, TLI 0.88, RMSEA 0.07). Alpha was 0.88 with the four factors between 0.66 and 0.87, and the two-week retest was 0.87. Scores correlated with an eating disorder inventory (r = 0.65), emotional eating (0.28) and negative body self (0.75), and went the other way with self-control (r = -0.31).
What a high score means, and does not
A total at or above the threshold means the pattern deserves a professional conversation, not that an eating disorder is present. The scale misses people who restrict without reporting the thoughts it asks about, and the six-point format performs less well than ideal according to Rasch analysis. If eating, weight or body image is taking up a large part of your day, or if you are bingeing, purging or exercising to compensate, that is reason enough to talk to someone regardless of your score. This page is not a diagnosis and does not replace one.