Composition monitoring and control play an essential role during a batch distillation cycle, but on-line composition analyzers are expensive, difficult to maintain and give delayed responses. Considering the need and lack of a stochastic estimator for batch distillation columns, a discrete extended Kalman filter (EKF) for binary and multicomponent systems has been developed and tested. The aim of the EKF was to provide reliable and real-time column composition profiles from few temperature measurements and easily available information. Accurate composition estimates and fast convergence were obtained, and the EKF has confirmed its ability to incorporate the effects of noise (from both measurement and modeling). The number of sensors and the observation frequency have shown to be important variables in the design of the EKF, especially for systems with fast dynamics.