Availability of personal computer programs for control of drug regimens has stimulated interest in modeling population pharmacokinetics. In this study, we found parameter values for gentamicin in two infant populations with low birth weights. The models were developed by use of a parametric (i.e., standard two-stage algorithm) and with a new nonparametric expected maximum algorithm. Data for the two populations (i.e., infants less than or equal to 31 weeks' and greater than 31 but less than or equal to 34 weeks' gestational age) were obtained from infants admitted to the University of Texas Medical Branch intensive care nursery between August 1, 1988, and July 31, 1989. The new nonparametric method was demonstrated to be not only the equal of the standard two-stage method for population modeling but better, especially in use of sparse data sets (e.g., single serum levels). It also obviates the need for selecting proper starting conditions for the least-squares fitting procedure used in the standard two-stage method.