Seismic reservoir characterization and monitoring require the knowledge of seismic wave velocities and their dependencies on reservoir properties and production-induced changes. In heavy-oil saturated rocks at cold temperatures, due to the nonzero shear rigidity of the fluid, the saturated shear modulus is higher than the dry shear modulus and, consequently, the observed P- and S-wave velocities are higher than Gassmann’s predicted velocities. Appropriate modeling of the saturated shear modulus can greatly enhance the accuracy of quantitative interpretation of spatial fluid saturation and temperature distribution within a reservoir undergoing thermal production. Using a well-log data set of an Athabasca heavy-oil play and measured oil viscosities from core samples, we estimate fluid viscosity, shear modulus, and the American Petroleum Institute (API) gravity logs by training a neural network (NNT) with available well logs. We also estimate the dry shear modulus of heavy-oil saturated rocks using an NNT approach after modeling the pressure variations within the reservoir. Our empirical model uses the apparent shear modulus of the oil, its saturation, porosity, and dry shear modulus to estimate the saturated shear modulus of the rock. We calibrate the model to ultrasonic lab measurements. Available literature data support the validity of the model and show the improved performance compared to the Ciz and Shapiro model. The range of applicability of the model is defined mathematically, and the behavior of the model with respect to the input parameters is examined through sensitivity analyses.