Use of a state-space grid form of the Bayes’ rule is advocated over the popular extended Kalman filter (EKF) for state estimation given nonnormal probability distributions or nonlinear processes, as is common in batch processes. The EKF cannot account for a nonnormal initial distribution or for the distortion of a normal distribution through a nonlinear model, and is subject to relatively large error due to linearization around the expected state value. A grid discretization of the state space allows for arbitrary initial distributions, faithful processing of the distribution through nonlinear models, and smaller error due to dynamic linearization that is performed around the state value at each grid point. The probability-grid filter (PGF) used here thus exhibits faster convergence to better estimates and narrower distributions, reduced sensitivity to tuning parameters, and extenuated susceptibility to divergence. A simulation example of a typical batch reactor is used to demonstrate the benefits of the PGF over the EKF when the process is significantly nonlinear or when the a priori knowledge translates into a nonnormal initial distribution. The benefits of the PGF come at the cost of increased computational effort.