Background Allogeneic islet transplantation is a validated therapy in type 1 diabetes; however, there is decline of transplanted islet graft function over time and the mechanisms underlying this decline are unclear. We evaluated the distinct association between primary graft function (PGF) and 5-year islet transplantation outcomes. Methods In this retrospective, multicentre, observational cohort study, we enrolled all patients from the Collaborative Islet Transplant Registry who received islet transplantation alone (ITA recipients) or islet-after-kidney transplantation (IAK recipients) between Jan 19, 1999, and July 17, 2020, with a calculable PGF (exposure of interest), measured 28 days after last islet infusion with a validated composite index of islet graft function (BETA-2 score). The primary outcome was cumulative incidence of unsuccessful islet transplantation, defined as an HbA1c of 7·0% (53 mmol/mol) or higher, or severe hypoglycaemia (ie, requiring third-party intervention to correct), or a fasting C-peptide concentration of less than 0·2 ng/mL. Secondary outcomes were graft exhaustion (fasting C-peptide <0·3 ng/mL); inadequate glucose control (HbA1c ≥7·0% [53 mmol/mol] or severe hypoglycaemia); and requirement for exogenous insulin therapy (≥14 consecutive days). Associations between PGF and islet transplantation outcomes were explored with a competing risk analysis adjusted for all covariates suspected or known to affect outcomes. A predictive model based on PGF was built and internally validated by using bootstraps resampling method. Findings In 39 centres worldwide, we enrolled 1210 patients with a calculable PGF (of those without missing data, mean age 47 years [SD 10], 712 [59·5%] were female, and 865 (97·9%) were White), who received a median of 10·8 thousand islet-equivalents per kg of bodyweight (IQR 7·4–13·5). 986 (82·4%) were ITA recipients and 211 (17·6%) were IAK recipients. Of 1210 patients, 452 (37·4%) received a single islet infusion and 758 (62·6%) received multiple islet infusions. Mean PGF was 14·3 (SD 8·8). The 5-year cumulative incidence of unsuccessful islet transplantation was 70·7% (95% CI 67·2–73·9), and was inversely and linearly related to PGF, with an adjusted subhazard ratio (sHR) of 0·77 (95% CI 0·72–0·82) per 5-unit increase of BETA-2 score (p<0·0001). Secondary endpoints were similarly related to PGF. The model-adjusted median C-statistic values of PGF for predicting 5-year cumulative incidences of unsuccessful islet transplantation, graft exhaustion, inadequate glucose control, and exogenous insulin therapy were 0·70 (range 0·69–0·71), 0·76 (0·74–0·77), 0·65 (0·64–0·66), and 0·72 (0·71–0·73), respectively. Interpretation This global multicentre study reports a linear and independent association between PGF and 5-year clinical outcomes of islet transplantation. The main study limitations are its retrospective design and the absence of analysis of complications. Funding Public Health Service Research, National Institutes of Health, Juvenile Diabetes Research Foundation International, Agence National de la Recherche, Fondation de l'Avenir, and Fonds de Dotation Line Renaud-Loulou Gasté.