The assumptions made in deriving models of soluble gas exchange used for continuous estimation of pulmonary perfusion are critically examined. Comparisons are made between estimation algorithms based on simple and more complex models. The more complex model includes a general treatment of tidal breathing, an inhomogeneous lung comprising multiple distensible compartments, and nonlinearities due to multiple-gas effects. The results show that sensitivity of perfusion estimates to errors inherent in simple linear models. These errors can invalidate the estimates under realistic physiological conditions. Concentration and multiple-gas effects, for example, can cause substantial estimation errors. Large ventilations relative to lung volume can also lead to errors. These simulations can be used to delimit the conditions under which the estimates can be considered reliable. A further set of simulations is used to assess the sensitivity of perfusion estimates to errors in the assumed values of unknown or approximately known model parameters. Inadequate specification of the compartmental structure of the lung (distributions of ventilation/perfusion and ventilation/volume) can cause large estimate errors. Precise estimates of lung tissue volume do not appear to be necessary. These results are important for the practical application of soluble gas methods for pulmonary perfusion determination. In both sets of simulations, it is shown that parameter estimate accuracy must be confirmed independently of goodness-of-fit criteria. Close agreement between predicted and observed end-tidal concentrations does not ensure accurate perfusion estimates.