In this paper, we aim to efficiently estimate parameters of biological systems. Biological systems, and more specifically biochemical networks, are usually modeled by Ordinary Differential Equations (ODEs). Instead of working directly on the ODE models, our approach relies on approximating biological systems with discrete probabilistic models, more specifically Dynamic Bayesian Networks (DBNs). The discrete approximation is used to efficiently estimate the parameters of the system by relying on Bayesian inference instead of classical optimization methods that fail due to the high dimensionality and number of parameters encountered in this type of systems. While more efficient than classical approaches, our method is also limited by dimensionality. To mitigate the dimensionality issue as much as possible, we tackle the problem of model reduction of biological systems, and the implications it has on our parameter estimation method. More specifically, we discuss an exact model reduction method that allows us to simplify the system without losing information, and we adapt the discrete models and parameter estimation methods to this reduction method. We illustrate our methods on some concrete case studies.