Information about the concentration of unbound paclitaxel over time following administration of albumin-bound paclitaxel (nab-paclitaxel) and its proportion to total paclitaxel in plasma remains unavailable. The aim of this study was to establish a population pharmacokinetic (PK) model to evaluate the dynamic relationship between total and unbound concentrations of paclitaxel in Chinese patients with metastatic breast cancer. A total of 653 concentrations corresponding to total and 334 to unbound paclitaxel were analyzed in 24 subjects who received two cycles of a single nab-paclitaxel dose of 260 mg/m2/cycle. The time course of the fraction of unbound paclitaxel was analyzed by non-linear mixed-effects modeling and Monte Carlo simulation. A mechanism-based model incorporating linear dissociation of nab-paclitaxel and saturated binding of free paclitaxel to plasma components was established to describe the relationship between total and unbound paclitaxel. Dissociation coefficient of nab-paclitaxel was estimated to be 4.60%, while a maximum unbound fraction value of 2.76% was reached at the end of infusion due to nonlinear binding kinetics of unbound paclitaxel. This is the first study to determine relationship between total and unbound nab-paclitaxel concentrations using a semi-mechanical population PK model, revealing that unbound paclitaxel fraction was expressed as a function of nab-paclitaxel concentration.