Partial least squares is a data-driven modeling technique that has been utilized for process monitoring in a variety of industrial processes. This paper develops a novel online partial least squares approach (evolving PLS) and compares it with an existing online PLS technique (global PLS). Both methods are applied to an industrial fed-batch mammalian cell culture process, where process variables are used to predict a key quality variable, product titer. Fault detection and diagnosis are performed using PLS models and statistical metrics. This new detection approach was able to recognize a variety of faults during online monitoring. (C) 2008 Elsevier Ltd. All rights reserved.