OBJECTIVE We previously reported that 2 years of costimulation modulation with abatacept slowed decline of β-cell function in recent-onset type 1 diabetes (T1D). Subsequently, abatacept was discontinued and subjects were followed to determine whether there was persistence of effect. RESEARCH DESIGN AND METHODS Of 112 subjects (ages 6-36 years) with T1D, 77 received abatacept and 35 received placebo infusions intravenously for 27 infusions over 2 years. The primary outcome-baseline-adjusted geometric mean 2-h area under the curve (AUC) serum C-peptide during a mixed-meal tolerance test (MMTT) at 2 years-showed higher C-peptide with abatacept versus placebo. Subjects were followed an additional year, off treatment, with MMTTs performed at 30 and 36 months. RESULTS C-peptide AUC means, adjusted for age and baseline C-peptide, at 36 months were 0.217 nmol/L (95% CI 0.168-0.268) and 0.141 nmol/L (95% CI 0.071-0.215) for abatacept and placebo groups, respectively (P = 0.046). The C-peptide decline from baseline remained parallel with an estimated 9.5 months' delay with abatacept. Moreover, HbA1c levels remained lower in the abatacept group than in the placebo group. The slightly lower (nonsignificant) mean total insulin dose among the abatacept group reported at 2 years was the same as the placebo group by 3 years. CONCLUSIONS Costimulation modulation with abatacept slowed decline of β-cell function and improved HbA1c in recent-onset T1D. The beneficial effect was sustained for at least 1 year after cessation of abatacept infusions or 3 years from T1D diagnosis.