We propose a new method (MDQC) to identify potentially low-quality arrays. Its advantage is that it has a clear statistical foundation, it uses the correlation structure of the various QC measures, it is easy to apply, and it is computationally light-weight. These properties make MDQC a useful diagnostic technique suitable for large datasets. MDQC performs a robust multivariate analysis of the quality measures provided in the QC report while taking into account their correlation structure. More precisely, the method first identifies the typical quality measures of valid arrays using robust estimators of the center and correlation structure. It then uses the MD based on these estimators to flag arrays with quality measures that are far from those of valid ones. We show that a multivariate analysis gives substantially richer information than the inspection of individual measures in isolation. Moreover, the method gives a simple way to compare the quality across arrays that is useful to rank them according to their quality and to flag those likely to be defective. Finally, we show that computing these distances on subsets of the quality measures in the report, instead of on all of them, may increase the method's ability to detect unusual arrays. In our case studies, we find that the a priori grouping method and the global PCA identify almost the same set of multivariate outliers. However, using the a priori method, the interpretability of the groups may be used to provide useful information about the likely source of potential quality problem.