A scheme has been developed for state and parameter estimation in an industrial-scale fixed-bed reactor with decaying catalysts where phthalic anhydride is produced by the partial oxidation of o-xylene. In order to account for reversible and irreversible catalyst deactivation, an activity profile with unknown functional form was incorporated into the reactor model which included a simplified reaction network. The distributed parameter reactor model was lumped to a differential and algebraic system with the use of orthogonal collocation on finite elements. The optimal estimation of the reactor state was formulated as a nonlinear programme based on a moving horizon and solved by a sequential strategy. Using four real-time temperature measurements, the dynamic behaviour of the reactor was excellently tracked, and the catalyst activity profile was accurately inferred. The computing time required for the estimation is longer than that for an optimal filtering method using a mechanistic deactivation model.