Background Total knee replacement (TKR) or unicompartmental knee replacement (UKR) are options for end-stage osteoarthritis. However, comparisons between the two procedures are confounded by differences in baseline characteristics of patients undergoing either procedure and by insufficient reporting of endpoints other than revision. We aimed to compare adverse outcomes for each procedure in matched patients. Methods With propensity score techniques, we compared matched patients undergoing TKR and UKR in the National Joint Registry for England and Wales. The National Joint Registry started collecting data in April 1, 2003, and is continuing. The last operation date in the extract of data used in our study was Aug 28, 2012. We linked data for multiple potential confounders from the National Health Service Hospital Episode Statistics database. We used regression models to compare outcomes including rates of revision, revision/reoperation, complications, readmission, mortality, and length of stay. Findings 25 334 UKRs were matched to 75 996 TKRs on the basis of propensity score. UKRs had worse implant survival both for revision (subhazard ratio [SHR] 2·12, 95% CI 1·99–2·26) and for revision/reoperation (1·38, 1·31–1·44) than TKRs at 8 years. Mortality was significantly higher for TKR at all timepoints than for UKR (30 day: hazard ratio 0·23, 95% CI 0·11–0·50; 8 year: 0·85, 0·79–0·92). Length of stay, complications (including thromboembolism, myocardial infarction, and stroke), and rate of readmission were all higher for TKR than for UKR. Interpretation In decisions about which procedure to offer, the higher revision/reoperation rate of UKR than of TKR should be balanced against a lower occurrence of complications, readmission, and mortality, together with known benefits for UKR in terms of postoperative function. If 100 patients receiving TKR received UKR instead, the result would be around one fewer death and three more reoperations in the first 4 years after surgery.