A new nonparametric expectation maximisation (NPEM) algorithm for the estimation of population pharmacokinetic parameter values was evaluated. The algorithm, in the form of a personal computer program, was used to compute population pharmacokinetic parameter densities of gentamicin in a group of 9 patients with indicators of malnutrition. The 1-compartment parameter values for clearance (CL), volume of distribution (Vd) and elimination rate constant (k) were compared with values generated using a standard 2-stage (STS) approach. NPEM was used with a full data set (72 gentamicin concentrations) and a sparse data set (only peak and trough concentrations for each patient; 18 in total). There were no differences in parameter value estimations between the STS and NPEM with all the data (p greater than 0.05) or with the sparse data (p greater than 0.05). Mean parameter value estimates from the STS and NPEM (with sparse data) were used as a priori data sets in the USC*PACK gentamicin Bayesian program to predict concentrations in 8 subsequent patients with similar indicators of malnutrition. There were no differences in predicted gentamicin concentrations between STS (3.75 +/- 2.06 mg/L) and NPEM (3.75 +/- 2.17 mg/L). NPEM was able to generate population pharmacokinetic parameter values for gentamicin in a defined population of patients using sparse routine clinical data. It was also shown to function with only a single data point per patient.