Beck Anxiety Inventory: 21 items about the body
Anxiety shows up in two places. One is what you worry about. The other is what your body does while you worry. The BAI was written to capture the second, and 14 of its 21 items are physical, which is why it reads differently from a worry questionnaire.
Beck, Epstein, Brown and Steer published it in 1988. The items were distilled from three earlier checklists holding 86 statements between them, cut to 47, then 37, then the final 21. Each is rated 0 to 3 for the past week, including today, and the total runs from 0 to 63.
Bands
| Total | Band |
|---|---|
| 0-7 | Minimal, clinically non-significant |
| 8-15 | Mild |
| 16-25 | Moderate |
| 26-63 | Severe |
A second set of boundaries circulates in the rheumatology literature (0 to 9, 10 to 18, 19 to 29, 30 to 63), so a score quoted without its source can look like it belongs to a different band. Sixteen is the cutoff most clinics use for moderate anxiety, with sensitivity and specificity both reported above 80%. The Chinese manual recommends 16 to 20, where sensitivity runs .83 to .90 and specificity .84 to .89.
Reliability
Internal consistency was .93 in the original psychiatric outpatient sample. A meta-analysis of 117 studies put it at .91 in clinical samples and .91 in non-clinical ones. A Korean validation of 1,157 people reported .956, and Chinese versions range from 0.90 to 0.94 in general samples and 0.91 to 0.96 in anxiety patients.
Retest stability depends heavily on the interval, and the spread is worth seeing: .93 over one week, .84 in the Korean version, .62 over seven weeks, and .66 and .65 in the meta-analysis for clinical and non-clinical samples. Anxiety symptoms genuinely move, so a weaker long-interval correlation is not a defect in the scale.
What it measures, and what it does not
Convergent correlations are moderate to strong: .51 with the Hamilton Anxiety Rating Scale, .58 with the trait scale of the STAI and .47 with its state scale, .81 with the anxiety dimension of the SCL-90, .776 with the GAD-7, .724 with the Anxiety Sensitivity Index-3, .640 with a fear questionnaire and .518 with the Penn State Worry Questionnaire.
The scale was built to separate anxiety from depression, and that claim has worn thin. Its correlation with depression measures now runs as high as its correlation with anxiety measures: .48 with the BDI in the original sample of 160 outpatients, .61 in a non-clinical sample, .796 with the BDI-II in the Korean data, .769 with the PHQ-9 and .747 with the CES-D. Pooled across 109 studies and 28,533 people, the correlation between the BAI and the BDI is 0.59. In the original validation the anxiety group (n = 82, median 24) did score above the depression group (n = 30, median 13), with about 25% of the anxiety group above the depression group's highest score. In the Korean sample the pattern reversed: depressed patients averaged 15.12, anxious patients 11.90.
Two further limits matter. The somatic emphasis inflates scores in people with medical conditions, in older adults and possibly in East Asian samples, where distress is more often expressed through the body. And the scale says nothing about worry or rumination, so a low total does not rule out cognitive anxiety.
Screening performance
A 2025 Cochrane review pooled 14 studies and 6,232 people at a cutoff of 16. In a group of 1,000 where 270 genuinely have an anxiety disorder, 241 would screen positive and 95 of those would be false positives, while 124 of the 759 who screen negative would actually have the disorder. The review concluded that the evidence for the BAI as a screening tool is not strong enough, and that shorter questionnaires are worth considering first. A Korean study of 1,157 people with structured interview diagnoses put the best cutoff at 8, with sensitivity .750 and specificity .740 and an area under the curve of 0.928. The original 15 to 16 threshold caught only about half the cases in that sample.
For tracking change over time, the publisher reports that a shift of about 9 points is needed to reach statistical significance at the 90% confidence level.
Practical notes
Reference means help with placement. Non-clinical samples average 11.08 (SD 9.10) for students, 8.89 (SD 7.30) for medical students and 7.78 (SD 5.65) for non-student adults. Among diagnosed groups, panic disorder without agoraphobia averaged 28.81 (SD 13.46), panic disorder with agoraphobia 27.27 (SD 13.11), obsessive-compulsive disorder 21.69 (SD 12.42), generalised anxiety disorder 18.83 (SD 9.08) and social phobia 17.77 (SD 11.64). Women with anxiety disorders score roughly 4 points higher than men.
The BAI is copyrighted and published by Pearson, so the item wording is licensed rather than free. A total is not a diagnosis. If your score sits at 16 or above, or if physical symptoms are getting in the way of daily life, that is worth raising with a clinician. This page is not a diagnosis and does not replace one.