A numerical study on 2D natural convection in cylindrical cavities during the sterilization of liquid foods was performed. The mathematical model was established on momentum and energy balances and predicts both the heating dynamics of the slowest heating zone (SHZ) and the lethal rate achieved in homogeneous liquid canned foods. Two sophistication levels were proposed in viscosity modelling: 1) considering average viscosity and 2) using an Arrhenius-type model to include the effect of temperature on viscosity. The remaining thermodynamic properties were kept constant. The governing equations were spatially discretized via orthogonal collocation (OC) with mesh size of 25 × 25. Computational simulations were performed using proximate and thermodynamic data for carrot-orange soup, broccoli-cheddar soup, tomato puree, and cream-style corn. Flow patterns, isothermals, heating dynamics of the SHZ, and the sterilization rate achieved for the cases studied were compared for both viscosity models. The dynamics of coldest point and the lethal rate F0 in all food fluids studied were approximately equal in both cases, although the second sophistication level is closer to physical behavior. The model accuracy was compared favorably with reported sterilization time for cream-style corn packed at 303 × 406 can size, predicting 66 min versus an experimental time of 68 min at retort temperature of 121.1 ℃.