Escalation therapy with mitoxantrone (MX) in highly active multiple sclerosis is limited by partially dose-dependent side-effects. Predictors of therapeutic response may result in individualized risk stratification and MX dosing. ATP-binding cassette-transporters ABCB1 and ABCG2 represent multi-drug resistance mechanisms involved in active cellular MX efflux. Here, we investigated the role of ABC-gene single nucleotide polymorphisms (SNPs) for clinical MX response, corroborated by experimental in vitro and in vivo data. Frequencies of ABCB1 2677G>T, 3435C>T and five ABCG2-SNPs were analysed in 832 multiple sclerosis patients (Germany, Spain) and 264 healthy donors. Using a flow-cytometry-based in vitro assay, MX efflux in leukocytes from individuals with variant alleles in both ABC-genes (designated genotype ABCB1/ABCG2-L(ow), 22.2% of patients) was 37.7% lower than from individuals homozygous for common alleles (ABCB1/ABCG2-H(igh), P < 0.05, 14.8% of patients), resulting in genotype-dependent MX accumulation and cell death. Addition of glucocorticosteroids (GCs) inhibited MX efflux in vitro. ABC-transporters were highly expressed in leukocyte subsets, glial and neuronal cells as well as myocardium, i.e. cells/tissues potentially affected by MX therapy. In vivo significance was further corroborated in experimental autoimmune encephalomyelitis in Abcg2–/– animals. Using a MX dose titrated to be ineffective in wild-type animals, disease course and histopathology in Abcg2–/– mice were strongly ameliorated. Retrospective clinical analysis in MX monotherapy patients (n = 155) used expanded disability status scale, relapse rate and multiple sclerosis functional composite as major outcome parameters. The clinical response rate [overall 121 of 155 patients (78.1%)] increased significantly with genotypes associated with decreasing ABCB1/ABCG2-function [ABCB1/ABCG2-H 15/24 (62.5%) responders, ABCB1/ABCG2-I(ntermediate) 78/98 (79.6%), ABCB1/ABCG2-L 28/33 (84.8%), exact Cochran-Armitage test P = 0.039]. The odds ratio for response was 1.9 (95% CI 1.0–3.5) with each increase in ABCB1/ABCG2 score (from ABCB1/ABCG2-H to –I-, and –I to –L). In 36 patients with severe cardiac or haematological side effects no statistically relevant difference in genotype frequency was observed. However, one patient with biopsy proven cardiomyopathy only after 24 mg/m2 MX exhibited a rare genotype with variant, partly homozygous alleles in 3 ABC-transporter genes. In conclusion, SNPs in ABC-transporter genes may serve as pharmacogenetic markers associated with clinical response to MX therapy in multiple sclerosis. Combined MX/GC-treatment warrants further investigation.