Polymer process control is limited by a lack of observability of the distributed and transient polymer states. Three simulations of varying complexity are validated for on-line simulation of an injection molding process with a two drop hot runner system to predict the state of the polymer melt in real time and thereby improve product quality in situ. The simplest simulation is a Newtonian model, which predicts flow rates given the inlet and outlet pressures. An intermediate non-Newtonian and nonisothermal simulation utilizes a modified Ellis model that expresses the viscosity as a function of the shear stress in which the modeling of the heat transfer utilizes a Bessel series expansion to include effects of heat conduction, heat convection, and internal shear heating. A numerical simulation was also developed that utilizes a hybrid finite difference and finite element scheme to simultaneously solve the mass, momentum, and heat equations. Numerical verification indicates that the flow rate predictions of the described simulations compare well with the results from a commercial mold filling simulation. However, empirical validation utilizing a design of experiments indicates that the described analyses are qualitatively useful, but do not possess sufficient accuracy for quantitative process and quality control. Specifically, off-line validation using optimal transducer calibration with well characterized materials provided a coefficient of regression, R-2, of similar to 0.8. However, blind validation with previously untested materials and no transducer re calibration provided a regression coefficient of similar to 0.4. While the direction of the main effects was usually correct, the magnitudes of the effects were frequently outside the confidence interval of the observed behavior. Several sources of variance are discussed, including sensor calibration, constitutive modeling of the polymer melt, and numerical analysis.