Most gene expression profiling studies of mesothelioma have been based on relatively small sample numbers, limiting their statistical power. We did Affymetrix U133A microarray analysis on 99 pleural mesotheliomas, in which multivariate analysis showed advanced-stage, sarcomatous histology and P16/CDKN2A homozygous deletion to be significant independent adverse prognostic factors. Comparison of the expression profiles of epithelioid versus sarcomatous mesotheliomas identified many genes significantly overexpressed among the former, including previously unrecognized ones, such as uroplakins and kallikrein 11, both confirmed by immunohistochemistry. Examination of the gene expression correlates of survival showed that more aggressive mesotheliomas expressed higher levels of Aurora kinases A and B and functionally related genes involved in mitosis and cell cycle control. Independent confirmation of the negative effect of Aurora kinase B was obtained by immunohistochemistry in a separate patient cohort. A role for Aurora kinases in the aggressive behavior of mesotheliomas is of potential clinical interest because of the recent development of small-molecule inhibitors. We then used our data to develop microarray-based predictors of 1 year survival; these achieved a maximal accuracy of 68% in cross-validation. However, this was inferior to prognostic prediction based on standard clinicopathologic variables and P16/CDNK2A status (accuracy, 73%), and adding the microarray model to the latter did not improve overall accuracy. Finally, we evaluated three recently published microarray-based outcome prediction models, but their accuracies ranged from 63% to 67%, consistently lower than reported. Gene expression profiling of mesotheliomas is an important discovery tool, but its power in clinical prognostication has been overestimated.