The absence of a standardized definition for GVHD flares and data on its clinical course are significant concerns. We retrospectively evaluated 968 patients across 23 Mount Sinai Acute GVHD International Consortium (MAGIC) transplant centers who achieved complete response (CR) or very good partial response (VGPR) within 4 weeks of treatment. The cumulative incidence of flares within 6 months was 22% and flares were associated with a higher risk of non-relapse mortality (NRM) (adjusted Hazard Ratio [aHR] 4.84 [95% CI, 3.19-7.36], P < 0.001). Compared to the initial GVHD, flares were more severe (Grades III/IV: 41% vs. 16%, P < 0.001) and had more frequent lower gastrointestinal (LGI) involvement (55% vs. 32%, P < 0.001). At CR/VGPR, elevated MAGIC biomarkers predicted the future occurrence of a flare, along with its severity and LGI involvement. In multivariate analyses, higher Ann Arbor (AA) biomarker scores at CR/VGPR were significant risk factors for flares (AA2 vs. AA1: aHR, 1.81 [95% CI, 1.32-2.48], P = 0.001; AA3 vs. AA1: aHR, 3.14 [95% CI, 1.98-4.98], P < 0.001), as were early response to initial treatment (aHR, 1.84 [95% CI, 1.21-2.80], P = 0.004) and HLA-mismatched unrelated donor (aHR, 1.74 [95% CI, 1.00-3.02], P = 0.049). MAGIC biomarkers also stratified the risk of NRM both at CR/VGPR and at the time of flare. We conclude that GVHD flares are common and carry a significant mortality risk. The occurrence of future flares can be predicted by serum biomarkers that may serve to guide adjustment and discontinuation of immunosuppression.