Accurate well-timed measurement of quality variables is essential to the successful monitoring and controlling of wastewater treatment systems. Because the measurements of these variables are difficult and often involve large time delays, predictive models for target quality variables have been widely considered. However, many microbial reactions and their interactions with the environment result in time dependent processes, making the development of bioprocess models difficult and time-consuming. In this paper, steady-state and dynamic predictive models based on multiple linear regression (MLR) and partial least squares (PLS) regression are presented. Water quality measurements and process information are used to develop models to predict biochemical oxygen demand (BOD) at the inlet and outlet of an aerated lagoon of a pulp and paper mill operated by International Paper of Brazil (IPB). The results show that linear steady-state and dynamic models are able to predict inlet and outlet BOD even for a complex process that has operational data limitations (imprecise measurements, a large number of missing values, etc.). A companion paper [Chem. Eng. J., submitted for publication] reports static and dynamic nonlinear models that were developed from the same 4 years of data using a neural network approach. Together, the two papers provide a well-documented application of linear and nonlinear empirical modeling techniques to an industrial case study. The modeling techniques are also valid for other types of industrial applications.