This paper reports the application of the extended Kalman filtering (EKF) technology to the design of a nonlinear observer used to predict the state of an experimental biochemical reactor. The plant is composed by a control unit and a 2.5 liter glass bioreactor where the β-galactosidase enzyme is produced from whey cheese and Kluyveromyces marxianus yeast. In order to implement the EKF, an accurate mathematical model, previously developed by the authors, was considered. The implemented continuous-discrete version of EKF is used to infer the lactose concentration from the measured values of the dissolved oxygen concentration in the medium. The results show that the nonlinear observer works well, being suitable to correct the non-reliable laboratory analysis data as well as to implement advanced control techniques.