The Berg Balance Scale, often shortened to the Berg scale, scores 14 balance tasks from 0 to 4, for a total out of 56 points. It is in the public domain, and the full scorer is right below.

The minimal detectable change is not constant: it is 4 points for a baseline score between 45 and 56, and up to 7 points between 25 and 34. The thresholds are detailed below.

Figures taken from the studies listed at the end of this page, every value carrying its source where it is written. The item labels come from the public-domain version.

What the Berg Balance Scale measures

The Berg Balance Scale scores fourteen tasks, from a simple sit-to-stand to single-leg stance, each from 0 to 4, for a total out of 56 points. It needs only a chair, a step, a ruler and a stopwatch, and no prior training is required to administer it.

What it describes very well has been quantified by a recent meta-analysis of 55 studies and 6,865 participants after stroke: for identifying a patient able to walk independently, the area under the curve reaches 0.917, with 88.0% sensitivity and 85.1% specificity. That is remarkable performance for a fifteen-minute test.

In other words, the scale is an excellent descriptor of current functional balance. Which says nothing about its ability to predict the future, and that is where the trouble lies.

Scoring the fourteen tasks

The fourteen labels reproduced in the scorer above come from the public-domain version. The scale is subject to no licence, no royalty and no prior registration, which sets it apart from most instruments in this field.

One point to know before recording it in French in the notes: no validated French version is published. A translation circulates, distributed by a Quebec institution, but its psychometric properties have not appeared. The rendering used here is therefore a working translation, and it must not be presented to the patient or in the notes as a validated version.

Reliability, on the other hand, is excellent and has never been seriously contested. The authors themselves report internal consistency above 0.83 in older residents and above 0.97 after stroke, with inter-rater reliability of ICC 0.98. A systematic review of 11 studies and 668 participants confirms it: 0.98 within raters, 0.97 between them.

The 45 cut-off, and what it misses

The cut-off of 45 out of 56 is the one everyone uses to talk about falls risk. A prospective study of 210 community-dwelling older people, with falls recorded monthly for a year, gives its real sensitivity.

It is 25% for the occurrence of at least one fall, and 45% for multiple falls. A cut-off that misses three fallers out of four is not a falls screening tool. The study is co-signed by Katherine Berg herself, which makes the finding hard to set aside.

The 2025 meta-analysis points the same way in another population: after stroke, the area under the curve is 0.704 for predicting falls, against 0.917 for identifying independent walking. The scale is therefore excellent at what it is not asked to do, and poor at what it is most often used for.

What the score does not tell you

It does not say whether a change is real until you know where the patient started. The minimal detectable change of the scale is not constant: measured in 118 subjects over 65 retested at 48 hours, it is 4 points for a baseline score between 45 and 56, 5 points between 35 and 44, 7 points between 25 and 34, and 5 points below 25.

That is the direct consequence of the ceiling effect: at the top of the range the items discriminate less, and measurement noise changes size depending on where you stand. A systematic review indeed finds values ranging from 2.8 to 6.6 points across studies.

Two methodological caveats to finish. The 1992 original publication carries no numeric coefficient in its abstract: citing it for an ICC is a frequent attribution error. And the 2025 meta-analysis, the most complete to date, reports a high risk of bias in 48 of its 55 studies.

Three administration pitfalls

Using the 45 cut-off as a falls test

Its 25% sensitivity makes it a tool that falsely reassures three times out of four. The score describes a level of balance, it does not screen for falls risk, and the falls assessment still has to be done separately.

Comparing two scores without looking at the band

A gain of 5 points is a real change for a patient starting at 30, and stays within noise for a patient starting at 50. The threshold is chosen from the baseline score, not once and for all.

Presenting the translation as validated

No French version has published psychometric properties. Using a translation is legitimate, presenting it as validated is not, and that precision belongs in the notes.

Frequently asked questions

What does the Berg Balance Scale measure?

It scores 14 balance tasks, each from 0 to 4, for a total out of 56 points. It describes the level of functional balance, in static positions as well as in transfers and movement.

What is the Berg Balance Scale cut-off?

The cut-off of 45 out of 56 is the most used, but its sensitivity for the occurrence of at least one fall is only 25% in community-dwelling older people. It describes a level, it does not screen for risk.

How much change counts?

It depends on the baseline score: 4 points between 45 and 56, 5 points between 35 and 44, 7 points between 25 and 34, and 5 points below 25.

Is there a validated French version?

No. No psychometric validation of a French version is published. A translation circulates and is used, but it must not be presented as validated.

Is the Berg Balance Scale free to use?

Yes. It is in the public domain: no royalty, no licence, no prior registration, and no training required to administer it.

Page written by Anthony Baillon, physiotherapist, co-founder of Physio Learning. The scale itself is by Katherine Berg et al. and is in the public domain: this page documents it and puts it to work, it does not claim authorship of it.