The yoga trial: two of three measures improvedEmerging evidence
In 118 adults aged 60 and over (average age 75) with a curve of at least 40 degrees, hour-long hatha classes 3 times a week for 24 weeks were compared with monthly lunch-time health talks. Compared with the talks group, the flexicurve angle improved by 4.4% (about 1.8 degrees) and a related index by 5%, but the kyphometer angle didn't change significantly, and neither did height, chair-stand times or quality of life.
Half the yoga group came to at least 80% of classes, and there were "No negative side effects". The researchers called it "developmental, not definitive" and suggested testing it in people whose spines are still flexible.
Sources 1, 10
Shoulder blades and reaching: no comparison groupMixed evidence
In 21 people from the trial, the shoulder blades sat 2.9% less far forward when sitting and a timed reaching task got 26.4% faster after 6 months, but there was no comparison group, and the highest point they could reach didn't change.
Source 12
Exercise in general: small changesEmerging evidence
Across 24 studies of exercise in people aged 45 and over with a rounded upper back, the curve improved by a small amount (moderate certainty), back-muscle endurance improved by about 10 seconds, and there were no serious side effects. In the best single trial, 6 months of physiotherapist-led spine strengthening reduced the curve by 3 degrees on X-ray compared with a health education group, without improving physical function. Older reviews agree that exercise can help a little, from studies of mixed quality.
Sources 13, 4, 14, 15, 16
Strengthening, not stretchingEmerging evidence
In 23 studies of 969 healthy people, strengthening changed spinal posture, most in the upper back and neck, while stretching on its own didn't, at moderate certainty. A 2025 consensus of 20 stretching researchers agreed that stretch training doesn't improve posture and that "strengthening weak muscles seems advisable to change posture". Applying this to yoga's back-strengthening poses is our inference, supported by the one yoga trial, which changed one curve measure by about 1.8 degrees.
Sources 2, 3, 1
A forward head: about 3 degrees, with very mixed resultsMixed evidence
In 10 trials of people with neck pain, exercise programmes improved the head-neck angle by about 3.4 degrees, at low to moderate certainty, but the results varied a great deal from trial to trial, and some trials compared one active treatment with another. A second 2026 review, of 13 trials, rated the evidence low to very low.
Sources 5, 17, 18
No trials of yoga for the head, shoulders or pelvisMixed evidence
We found no randomised trial of yoga for forward head posture, rounded shoulders or a tilted pelvis. The studies that exist are small: 42 women given 10 weeks of "cyclic yoga" (the summary gives no numbers), 28 people who all did yoga, and 56 women over 50 with no comparison group. In a comparison of 667 practitioners with 578 students, the practitioners had less pronounced back curves, but that can't show that yoga caused it, and the groups differed in age.
Sources 19, 20, 21, 22, 23, 24
Pain eased, posture didn't changeEmerging evidence
In 56 people with long-lasting neck pain, 12 yoga or Pilates classes reduced pain and disability, but "No improvements in range of movement or posture were found". In 150 office workers, hatha yoga eased neck and shoulder pain about as well as stretching. Feeling better doesn't depend on a change in posture.
Sources 25, 26
Yoga or strength training for posture?Mixed evidence
No trial has compared them. The yoga trial's change and the strengthening trial's 3 degrees were in different people and measured in different ways, so the numbers can't be set side by side.
Sources 1, 4
Is it safe?Emerging evidence
The yoga trial reported no negative side effects, and exercise trials for a rounded upper back reported no serious ones. The UK osteoporosis consensus links spinal fractures with "some yoga positions involving extreme spinal flexion", so with fragile bones, change the poses (below).
Sources 1, 13, 9