Exhaust Gas Recirculation (EGR) systems are used in internal combustion engines to reduce harmful emissions such as nitrogen oxides NOx. The accuracy of EGR mass flow estimation is critical for optimizing the performance of EGR systems. Traditionally, this has been achieved through the use of physical sensors, but there has been a growing interest in the use of virtual sensors as a viable alternative. This thesis discusses two different approaches to developing virtual sensors for EGR systems: physics-based modeling and neural network modeling. The experiments in this thesis aim to compare the accuracy and performance of these approaches in estimating the EGR mass flow rate.