Incidence on relapse and nonrelapse mortality (NRM) of chronic graft-versus-host disease (GVHD), per National Institutes of Health (NIH) criteria, is not well defined after reduced-intensity conditioning (RIC) regimens. We analyzed the association of chronic GVHD with the risk of relapse and NRM using Cox models in 177 consecutive patients who underwent transplantation for hematological malignancies after RIC. The cumulative incidence of chronic GVHD at 36 months was 74% when using Seattle's criteria compared with 54% with NIH consensus. In Cox model, NRM was significantly higher in patients with late-onset, persistent and recurrent acute GVHD (hazard ratio (HR): 6, 25 and 11; P = 0.014, P<0.0001, P<0.0001, respectively). The cumulative incidence of relapse was significantly decreased in patients with chronic GVHD compared with no GVHD group using either Seattle's or NIH criteria (HR 0.43 and 0.38; P = 0.022 and 0.016, respectively), whereas the presence of late-onset, persistent and recurrent acute GVHD was not associated with a decreased rate of relapse (HR: not significant, 0.70 and 0.71; P = not significant, P = 0.73 and P = 0.54, respectively). Chronic GVHD per NIH consensus definition is associated with the graft-versus-tumor effect, whereas all forms associated with acute features beyond day 100 are associated with NRM.