Identifying step changes in historical process and controller output variables can lead to improved process understanding and fault resolution in control system performance analysis. This paper describes an application of Bayesian methods in the search for statistically significant temporal segmentations in the data collected by a cross directional (CD) control system in an industrial web forming process. CD control systems give rise to vector observations which are often transformed through orthogonal bases for control and performance analysis. In this paper two models which exploit basis function representations of vector times series data are segmented. The first of these is a power spectrum model based on the asymptotic Chisquared approximation which allows large data sets to be processed. The second approach, more capable of detecting small changes, but as a result is more computationally demanding, is a special case of the multivariate linear model. Given the statistical model of the data, inference regarding the number and location of the change-points is based on numerical Bayesian methods known as Markov chain Monte Carlo (MCMC). The methods are applied to real data and the resulting segmentation relates to real process events.