PURPOSE: To create a model based on known mechanisms of paclitaxel distribution that could describe the pharmacokinetics (PK) of total and unbound plasma concentrations, as well as blood concentrations. In addition, to investigate the relationship between exposure, based on unbound and total concentrations, and neutropenia. PATIENTS AND METHODS: Paclitaxel and Cremophor EL (CrEL) concentrations were obtained from 23 female and three male patients (50 courses in total) with different cancer types that received paclitaxel (Taxol; Bristol-Myers Squibb Co, Princeton, NJ) (135 to 225 mg/m²) as 3- or 24-hour intravenous infusions. Seven of the patients received combination therapy with doxorubicin or cisplatin. The population PK model was built to fit three types of data simultaneously: unbound, total plasma, and blood concentrations. The area under the curve, threshold, and general models were used to relate neutrophil survival fraction from 19 patients (29 courses in total) to exposure based on unbound and total plasma concentration, respectively. RESULTS: The PK model included a linear three-compartment model for unbound concentration, binding directly proportional to CrEL, linear and nonlinear binding to plasma proteins, and linear and nonlinear binding to blood cells. The threshold model best described the PK/pharmacodynamic (PD) relationship for total concentration. No distinction could be made between the models for unbound drug. CONCLUSION: Earlier PK models for paclitaxel have been empirical. This study shows that a mechanistic model can be used to describe the nonlinear PK of paclitaxel. There is an indication that the PK/PD relationship is not the same for unbound and total plasma concentrations.