Principal components regression (PCR) is applied to the dynamic inferential estimation of plant outputs from highly correlated data. A genetic algorithm (GA) approach is developed for the optimal selection of subsets from the available measurement variables, thereby providing a method of identifying nonessential elements. The theoretical link between principal components analysis (PCA) and state-space modelling is employed to identify a measurement equation involving the GA-selected subset, which is then used for inferential estimation of the omitted variables. These techniques are successfully demonstrated for the inferential estimation of outputs from a validated industrial benchmark simulation of an overheads condensor and reflux drum model (OCRD).