Aims: To model the biotransformation steps of 5-FU production from capecitabine and identify patient characteristics that may influence the drug disposition. Methods: Blood samples and demographic data were collected from two phase I studies in which adult patients received oral capecitabine for various malignancies. Capecitabine, 5′-deoxy-5-fluorocytidine (5′-DFCR), 5′-deoxy-5-fluorouridine (5′-DFUR) and 5-fluorouracile (5-FU) concentration-time data were analysed via a population approach using NONMEM. Results: Forty patients and 75 pharmacokinetic time-courses were available for analysis. Capecitabine pharmacokinetics was ascribed to a one compartment model from which 5′-DFCR, 5′-DFUR and 5-FU were sequentially produced. Capecitabine oral absorption was characterized by a rapid first order input (K_a=2.1 ± 0.3 hr⁻¹) with a lag time (0.28 ± 0.11 hr), but related inter-occasion (IOV) and inter-subject (ISV) variabilities for these parameters, 167% and 110%, indicated that this oral absorption was highly variable. The capecitabine CL (CL₁₀ = 218± 18 L/hr, ISV = 18%) and 5′-DFUR elimination rate constant (K₃₄ = 5.3 ± 2.0 hr⁻¹, ISV = 25%) were influenced by total bilirubin (BILT). The elimination rate constant of plasma 5-FU (K₄₀) was 66 ± 24 hr⁻¹ (ISV = 34%).The final pharmacokinetic model was validated using 2000 bootstrap runs and provided non-parametric statistics of the parameters (median, 2.5th and 97.5th percentiles). Conclusions: This study supported the possibility of modelling a complex sequential metabolic pathway which produces pharmacologicaly active compounds from a prodrug. Only BILT significantly influenced the pharmacokinetics but this effect was not considered as relevant for dosing adjustment.