We examined whether genetic polymorphisms (SNPs) in the capecitabine activation pathway and CDA enzymatic activity were associated with prognosis, benefit from capecitabine-containing treatment or capecitabine-related toxicities. The study population comprised 188 metastatic breast cancer patients of the ATX trial (EudraCT 2006-006058-83) randomized for first-line paclitaxel and bevacizumab with (ATX) or without capecitabine (AT). Cumulative capecitabine dose until grade ≥2 hand-foot syndrome or until first dose reduction were toxicity endpoints. We genotyped CDA c.-451C. >. T (rs532545), CDA c.-33delC (rs3215400) and CES2 c.-806C. >. G (rs11075646). CDA activity in baseline serum was measured with a spectrophotometric assay and values were analyzed using a median cut-off or as continuous variable. CDA c.-33delC was prognostic for overall survival (OS) independent of hormone receptor status. For the predictive analysis, progression-free survival benefit from ATX over AT was observed in patients with a CDA c.-33del/del or del/insC genotype, a CDA c.-451CC or CT genotype, and a CES2 c.-806CC genotype compared with their counterparts. There was a higher response rate for ATX over AT in patients with a CDA c.-451CT or TT genotype. Patients with high CDA enzymatic activity had more benefit from capecitabine, while this was marginally observed in the CDA low group. Toxicity endpoints were not associated with any candidate markers. In conclusion, CDA c.-33delC was associated with OS. Since particular SNPs in CDA and CES2 were associated with benefit from the addition of capecitabine to AT, their predictive value should be explored in a higher number of patients.