A physiologically based pharmacokinetic model for trichloroethylene (TCE) in rodents and humans was calibrated with published toxicokinetic data sets. A Bayesian statistical framework was used to combine previous information about the model parameters with the data likelihood, to yield posterior parameter distributions. The use of the hierarchical statistical model yielded estimates of both variability between experimental groups and uncertainty in TCE toxicokinetics. After adjustment of the model by Markov chain Monte Carlo sampling, estimates of variability for the animal or human metabolic parameters ranged from a factor of 1.5-2 (geometric standard deviation [GSDI). Uncertainty was of the same order as variability for animals and higher than variability for humans. The model was used to make posterior predictions for several measures of cancer risk. These predictions were affected by both uncertainties and variability and exhibited GSDs ranging from 2 to 6 in mice and rats and from 2 to 10 for humans. The recent development of a comprehensive physiologically based pharmacokinetic (PBPK) model of trichloroethylene (TCE) disposition and metabolism in mice, rats, and humans (1) offers us the opportunity to examine issues of variability and uncertainty for that solvent. In particular, uncertainties in prediction of various cancer dose metrics deserve to be computed, since they could be directly used as input for improved risk assessments.PBPK modeling provides a strong mechanistic basis for prediction of disposition and metabolism of toxicants. Yet much uneasiness remains with the use of these models in toxicology (2). Similarly, as discussed in a recent review and an accompanying commentary, PBPK modeling has not seen the development it promised for therapeutic compounds (3,4). The reason for this essentially lies in the lack of statistical methods for calibrating these models. Because of individual variability and uncertainty, many parameters are difficult to measure accurately even for inbred animal strains. Using input parameters or presenting results in the form of a single value can therefore be very misleading (5). In the absence of rigorous statistical treatment, inference presented by PBPK modeling is largely empirical, hypotheses are left unvalidated, and predictions lack realistic measures of uncertainty. This state of affairs is unfortunate, when considering the consequences (for public health and national welfare) of the decisions made using these models.Obviously, correct statistical treatment of PBPK models is difficult, since these are large nonlinear models with relatively small data sets and a high degree of uncertainty and biological variability (6). It is also essential to respect the fundamental specificity of PBPK models, i.e., their high prior information content, which they provide through the opportunity to use physiological information on parameter values. Yet, although several parameters have physiological meaning and a narrow range of possible values, others-often specific of the com...