The variation in product quality of industrial poly-ethylene terephthalate (PET) produced in a spindraw-winding process was related to on-line measured process variables. An experimental design in three process variables was used to introduce meaningful variation. Using standard multivariate data analysis techniques the relation between the variation of the product quality and the process variation could be modeled. The adequacy of these models was judged by comparing the modeling error to the known measurement error of the quality variables. It Could be established that the three process variables that were used in the experimental design are responsible for almost all variation in product quality. This shows the feasibility of the multivariate approach to systematically analyze the variation in product quality of industrial processes.