Objective: Glucagon-like peptide (GLP)-1 receptor agonist treatment improves β-cell function. In this study, we investigated whether the improvements are sustained during a 3-year treatment period. Research design and methods: Sixty-nine metformin-treated type 2 diabetes patients were randomised to the GLP1 receptor agonist, exenatide (EXE) twice daily (BID) or to insulin glargine (GLAR). β-cell function parameters were derived using the Mari model from standardised breakfast and lunch meals that were administered before treatment, and after 1 and 3 years of treatment. EXE was administered before breakfast. Results: Fifty-nine (EXE: n = 30; GLAR: n = 29) and thirty-six (EXE: n = 16; GLAR: n = 20) patients completed the meal at 1-and 3-year treatment respectively. After 3 years, groups had comparable glycaemic control (HbA1c: EXE 6.6 ± 0.2% and GLAR 6.9 ± 0.2%; P = 0.216). Compared with GLAR, at 1 and 3 years, EXE induced a stronger reduction in post-breakfast glucose concentrations (P < 0.001), with lower C-peptide levels (P < 0.001). Compared with GLAR, EXE increased insulin secretion at 8 mmol/L glucose throughout the study period (P < 0.01). Both treatments improved β-cell glucose sensitivity after 1-year treatment. However, only EXE treatment sustained this improvement for 3 years. No consistent changes in other β-cell parameters including rate sensitivity and potentiation were observed. Conclusions: Compared with GLAR, EXE improved the parameters of β-cell function, especially insulin secretion at 8 mmol/L glucose and β-cell glucose sensitivity, which was sustained during the 3-year treatment period.