Identifying relevant cytotoxicity genes using an ex-vivo lymphoblastoid cell line (LCLs) model has distinct advantages for pharmacogenomic discovery studies of cancer chemotherapy, including standardized treatment conditions, availability of large numbers of samples, and publicly available genotypic data. However, there is little proof of principal data to confirm the promise of this approach. One of the known targets of 5-fluorouracil (5-FU) treatment is thymidylate synthase (TYMS). We hypothesized that genetic variants in TYMS would alter cytotoxicity because of 5-FU treatment using a LCL model system. LCLs from the Centre d'Etude du Polymorphisme Humain (CEPH) pedigrees (N=427) were treated with eight concentrations of 5-FU for 72 h, and cytotoxicity was determined using an Alamar Blue assay. For a subset of the 30 International Haplotype Mapping project (HapMap) trios, genotype data for 46 single-nucleotide polymorphism (SNP) variants encompassing the TYMS gene were downloaded from the HapMap website. Using a mixed models approach, each SNP was tested for association to 5-FU cytotoxicity in the subset of HapMap trios. Putatively associated SNPs (P<0.01), were then genotyped in the remaining LCLs in the CEPH pedigrees and tested for association. Two intronic SNPs in TYMS (rs2847153 and rs2853533) were significantly associated (P<0.01) with 5-FU cytotoxicity in the HapMap subset using the mixed models approach. After genotyping these SNPs in the full CEPH pedigrees, the associations with cytotoxicity showed a more reliable significance (P<0.0005), as a result of the increase in sample size. These results highlight the importance of the TYMS gene variants in response to 5-FU treatment. Furthermore, they provide additional biological validation of the relevance of LCLs as a model for pharmacogenomic gene discovery in cancer chemotherapy.