Chronic GVHD (cGVHD) is the major cause of long-term morbidity after allogeneic hematopoietic stem cell transplantation (HSCT). There are no biomarkers that can consistently predict its occurrence. We aimed to evaluate whether numbers of antigen-presenting cell subsets in peripheral blood (PB) or serum chemokine concentrations are biomarkers of cGVHD occurrence. The study cohort comprised 101 consecutive patients undergoing allogeneic HSCT between January 2007 and 2011. cGVHD was diagnosed by both modified Seattle criteria and National Institutes of Health (NIH) criteria. Multicolor flow cytometry was used to determine the number of PB myeloid dendritic cells (DCs), plasmacytoid DCs, CD16+ DCs, and CD16+ and CD16- monocytes, as well as CD4+ and CD8+ T cells, CD56+ natural killer cells, and CD19+ B cells. Serum concentrations of CXCL8, CXCL10, CCL2, CCL3, CCL4, and CCL5 were measured by a cytometry bead array assay. At a median of 60 days after enrollment, 37 patients had developed cGVHD. Patients with cGVHD and those without cGVHD had comparable clinical characteristics. However, previous acute GVHD (aGVHD) was strongly correlated with later cGVHD (57% versus 24%, respectively; P=.0024). Each potential biomarker was screened for its association with cGVHD using the Mann-Whitney U test. Biomarkers that differed significantly (P < .05) between patients with cGVHD and those without cGVHD were analyzed by receiver operating characteristic (ROC) curve analysis to select the variables predicting cGVHD with an area under the ROC curve (AUC) >.5 and a P value <.05. A multivariate Fine-Gray model identified the following variables as independently associated with the risk of cGVHD: CXCL10 ≥592.650 pg/mL (hazard ratio [HR], 2.655; 95% confidence interval [CI], 1.298 to 5.433; P=.008), pDC ≥2.448/μL (HR, .286; 95% CI, .142 to .577; P < .001) and previous aGVHD (HR, 2.635; 95% CI, 1.298 to 5.347; P=.007). A risk score was derived based on the weighted coefficients of each variable (2 points each), resulting in the identification of 4 cohorts of patients (scores of 0, 2, 4, and 6). In a competing risk analysis to stratify patients at differing risk levels of cGVHD, the cumulative incidence of cGVHD was 9.7%, 34.3%, 57.7%, and 100% in patients with scores of 0, 2, 4, and 6, respectively (P < .0001). The score could nicely stratify the patients based on the risk of extensive cGVHD as well as NIH-based global and moderate to severe cGVHD. Based on ROC analysis, the score could predict the occurrence of cGVHD with an AUC of .791 (95% CI, .703 to .880; P < .001). Finally, a cutoff score ≥4 was identified as the optimal cutoff by Youden J index with a sensitivity of 57.1% and a specificity of 85.0%. A multiparameter score including a history of previous aGVHD, serum CXCL10 concentration, and number of pDCs in the PB at 3 months post-HSCT stratify patients at varying risk levels of cGVHD. However, the score needs to be validated in a much larger independent and possibly multicenter cohort of patients undergoing transplantation from different donor types and with distinct GVHD prophylaxis regimens.