Variability between individuals and samples is an important yet poorly studied phenomenon. One recently developed theoretical approach to the study of variability is the use of "population-based" models, where a large number of related models are studied instead of a single model. In particular, parameter sensitivity analysis can be performed by generating a large population of related models with varied parameters (such as different levels of ion channels), running simulations of those models to calculate physiological outputs (such as action potential duration, APD, and Ca2+ transient amplitude, Δ[Ca2+]i), then performing regression to relate the outputs to the parameters. This produces a matrix of sensitivity coefficients, each of which represents a quantitative and testable prediction of how changes in each parameter will affect each output. We have applied this method to a model of the rat ventricular myocyte that contains separate formulations for epicardial and endocardial cells. The analysis implicated the L-type Ca2+ current density (GCa) as an important determinant of Δ[Ca2+]i and APD in both cell types, with a reduction in GCa causing decreased APD and Δ[Ca2+]i. A reduction in transient outward K+ current density (Gto) causes an increase in APD in both cell types. Surprisingly, reduction of Gto was predicted to increase Δ[Ca2+]i in epicardial but not in endocardial cells. Experimental studies in progress confirm the surprising prediction that changes in Gto can indeed influence Ca2+ transients in rat ventricular myocytes. This work demonstrates how population-based modeling can generate counterintuitive and testable predictions, and also illustrates a quantitative framework to mechanistically relate measured differences in the levels of currents to behavioral variability between isolated cells.