Left ventricular pressure, P(t), and outflow Q(t), data were collected in anesthetized, open-chest rats and dogs. The data were used in a three-tiered validation procedure to evaluate 14 competing forms of elastance [E(t)]-resistance (R) left ventricle (LV) pump models. Competing models arose from considering two forms of parameterization of E(t), time variation versus no time variation in LV unstretched volume (Vd) and dependence versus no dependence of R on P(t) and isovolumic P(t). A descriptive test based on the normalized root-mean-square errors in the fit to P and, separately, in the fit to Q was used to distinguish between models. The best of the competing models was the one that treated Vd as a function of time and R as a constant. Models of this form fitted the data very well and were said to be descriptively valid. The best of the competing models were then asked to predict the observed responses to changes in afterload, preload, and prior-beat history. The models did not predict these conditions well and failed to pass the test for predictive validity. Additionally, the model parameters were judged not to represent their supposed physical homologs and, thus, failed the test for explanative validity. One cause for E(t)-R model failure was an inadequate representation of events at end systole. This deficiency was apparently due to not accounting for deactivation in the model. Other features may also be needed before a comprehensive LV model can be formulated. Identical conclusions were made from data from the rat and the dog.