BACKGROUND: The benefits of exercise in mild and moderate knee or hip osteoarthritis (OA) are apparent, but the evidence in severe OA is less clear. We recently reported that neuromuscular training was well tolerated and feasible in patients with severe primary hip or knee OA. The aims of this controlled before-and-after study were to compare baseline status to an age-matched population-based reference group and to examine the effects of neuromuscular training on patient-reported outcomes and physical function in patients with severe primary OA of the hip or knee. METHODS: 87 patients (60-77 years) with severe primary OA of the hip (n = 38, 55% women) or knee (n = 49, 59% women) awaiting total joint replacement (TJR) had supervised, neuromuscular training (NEMEX-TJR) in groups with individualized level and progression of training. A reference group (n = 43, 53% women) was included for comparison with patients' data. Assessments included self-reported outcomes (HOOS/KOOS) and measures of physical function (chair stands, number of knee bends/30 sec, knee extensor strength, 20-meter walk test) at baseline and at follow-up before TJR. Analysis of covariance (ANCOVA) was used for comparing patients and references and elucidating influence of demographic factors on change. The paired t-test was used for comparisons within groups. RESULTS: At baseline, patients reported worse scores than the references in all HOOS/KOOS subscales (hip 27-47%, knee 14-52%, of reference scores, respectively) and had functional limitations (hip 72-85%, knee 42-85%, of references scores, respectively). NEMEX-TJR (mean 12 weeks (SD 5.6) of training) improved self-reported outcomes (hip 9-29%, knee 7-20%) and physical function (hip 3-18%, knee 5-19%) (p < 0.005). Between 42% and 62% of hip OA patients, and 39% and 61% of knee OA patients, displayed a clinically meaningful improvement (≥15%) in HOOS/KOOS subscales by training. The improvement in HOOS/KOOS subscale ADL was greater for patients with knee OA than hip OA, while the improvement in subscale Sport/Rec was greater for patients with hip OA than knee OA. CONCLUSIONS: Both self-reported outcomes and physical function were clearly worse compared with the reference group. Neuromuscular training with an individualized approach and gradual progression showed promise for improving patient-reported outcomes and physical function even in older patients with severe primary OA of the hip or knee.