Bladder cancer is the fifth most commonly diagnosed malignancy in the United States and one of the most prevalent worldwide. It harbors a probability of recurrence of >50%; thus, rigorous, long-term surveillance of patients is advocated. Flexible cystoscopy coupled with voided urine cytology is the primary diagnostic approach, but cystoscopy is an uncomfortable, invasive procedure and the sensitivity of voided urine cytology is poor in all but high-grade tumors. Thus, improvements in noninvasive urinalysis assessment strategies would benefit patients. We applied gene expression microarray analysis to exfoliated urothelia recovered from bladder washes obtained prospectively from 46 patients with subsequently confirmed presence or absence of bladder cancer. Data from microarrays containing 56,000 targets was subjected to a panel of statistical analyses to identify bladder cancer-associated gene signatures. Hierarchical clustering and supervised learning algorithms were used to classify samples on the basis of tumor burden. A differentially expressed geneset of 319 gene probes was associated with the presence of bladder cancer (P < 0.01), and visualization of protein interaction networks revealed vascular endothelial growth factor and angiotensinogen as pivotal factors in tumor cells. Supervised machine learning and a cross-validation approach were used to build a 14-gene molecular classifier that was able to classify patients with and without bladder cancer with an overall accuracy of 76%. Our results show that it is possible to achieve the detection of bladder cancer using molecular signatures present in exfoliated tumor urothelia. Further investigation and validation of the cancer-associated profiles may reveal important biomarkers for the noninvasive detection and surveillance of bladder cancer.