Neoadjuvant chemotherapy with gemcitabine and carboplatin (GC) for invasive bladder cancer increases the chance of a radical response for a subset of patients, while other patients suffer from severe adverse drug reactions without any benefit. To establish a method for predicting the response to chemotherapy with GC, the expression profiles of biopsy samples from 37 advanced bladder cancers were analyzed using a microarray consisting of 38,500 genes or ESTs. Upon analysis of 9 'responder' and 9 'non-responder' tumors, 12 'predictive' genes were found to be significantly differentially expressed between the 'responder' and 'non-responder' groups, and a numerical prediction scoring system that clearly separated the responder group from the non-responder group was established. This system accurately predicted the drug responses of 18 of 19 additional test cases that were reserved from the original 37 cases. Moreover, a quantitative PCR-based prediction system was developed that may be feasible for routine clinical use, and the sensitivity of invasive bladder cancer to neoadjuvant chemotherapy with GC was able to be predicted by the expression patterns in this set of genes. Nearly 50% of patients treated with GC or methotrexate, vinblastine, doxorubicin and cisplatin (M-VAC) therapy have been reported to achieve complete or partial response to either of these therapies. When we applied this prediction system as well as the system for M-VAC, we expected that approximately 80% of the patients would achieve significant tumor shrinking (>60%) by selection of either the GC or M-VAC regimens. Our results suggest that the two prediction scoring systems lead to achievement of 'personalized therapy' for the treatment of invasive bladder cancer and should improve the quality of life for patients with this disease.