Improving haematopoietic cell transplantation outcomes by selection of an HLA-matched unrelated donor is best practice; however, donor selection by secondary characteristics is controversial. We studied 1271 recipients with haematological malignancies who underwent T-cell-depleted allografts and had complete data on HLA-matching status for six loci (HLA-A, -B, -C, -DRB1, -DQB1, -DPB1) and clinical outcome data. Five-year overall survival was 40.6%. HLA mismatching (at HLA-A, -B, -C, -DRB1, -DQB1) relative risk (RR) 1.22, 95% confidence interval (CI) 1.2-1.5, P = 0.033 for 1 mismatch and RR 1.46, 95% CI 1.1-1.9, P = 0.009 for >1 mismatch) and CMV mismatching (RR 1.37, 95% CI 1.2-1.6, P<0.001) were significantly associated with inferior survival. Donors aged <30 years showed a trend towards better survival. The multivariate model for mortality, combining CMV and HLA-match status, found an RR of 1.36 (95% CI 1.1-1.7, P = 0.003) for HLA matched/CMV mismatched, an RR of 1.22 (95% CI 0.99-1.5, P = 0.062) for HLA mismatched/CMV matched and an RR of 1.81 (95% CI 1.4-2.3, P = <0.001) for HLA/ CMV mismatched, compared with the HLA/CMV-matched recipients. These data suggest that HLA and CMV matching status should be considered when selecting unrelated donors and that CMV matching may abrogate the effect of an HLA mismatch.