Background: The modified Marsh and Schnider pharmacokinetic models for propofol consistently produce negatively and positively biased predictions in underweight patients, respectively. We aimed to develop a new pharmacokinetic model of propofol in underweight patients. Methods: Twenty underweight (BMI<18.5 kg m−2) patients aged 20–68 yr were given an i.v. bolus of propofol (2 mg kg−1) for induction of anaesthesia. Anaesthesia was maintained with a zero-order infusion (8 mg kg−1 h−1) of propofol and target-controlled infusion of remifentanil. Arterial blood was sampled at preset intervals. A population pharmacokinetic analysis was performed using non-linear mixed effects modelling. The time to peak effect (tpeak, maximally reduced bispectral index) was measured in 28 additional underweight patients receiving propofol 2 mg kg−1. Results: In total, 455 plasma concentration measurements from the 20 patients were used to characterise the pharmacokinetics of propofol. A three-compartment mammillary model well described the propofol concentration time course. BMI and lean body mass (LBM) calculated using the Janmahasatian formula were significant covariates for the rapid peripheral volume of distribution and for the clearance of the final pharmacokinetic model of propofol, respectively. The parameter estimates were as follows: V1(L)=2.02, V2(L)=12.9(BMI/18.5), V3(L)=139, Cl (L⋅min−1)=1.66(LBM/40), Q1 (L⋅min−1)=1.44, Q2 (L⋅min−1)=0.87+0.0189×(LBM−40). The median tpeak of propofol was 1.32 min (n=48). Conclusions: A three-compartment mammillary model can be used to administer propofol via target effect-site concentration-controlled infusion of propofol in underweight patients. Clinical trial registration: KCT0001760.