ArrayXpress uses an ANOVA approach with mixed model specification (Wolfinger et al., 2001. Journal of Computational Biology. 8: 625-637). Two steps of linear mixed models are applied. The normalization model is applied for global normalization at slide-level. The residual values from the normalization model are calculated for evaluating the model fitness. These residuals are then used to calculate true gene effects. The gene model is applied to allow for rigorous statistical inference on each gene. All random terms are assumed to be normal distributed and mutually independent within each model. ArrayXpress uses the SAS® Microarray Solution software for data analysis (www.sas.com).