The purpose of our study was to determine the most accurate analytic method to define in vitro chemosensitivity, using clinical response as reference standard in prospective clinical trial, and to assess accuracy of adenosine triphosphate-based chemotherapy response assay (ATP-CRA). Forty-eight patients with chemo-naïve, histologically confirmed, locally advanced or metastatic gastric cancer were enrolled for the study and were treated with combination chemotherapy of paclitaxel 175 mg/m(2) and cisplatin 75 mg/m(2) for maximum of six cycles after obtaining specimen for ATP-CRA. We performed the receiver operator characteristic curve analysis using patient responses by WHO criteria and ATP-CRA results to define the method with the highest accuracy. Median progression free survival was 4.2 months (95% confidence interval [CI]: 3.4-5.0) and median overall survival was 11.8 months (95% CI: 9.7-13.8) for all enrolled patients. Chemosensitivity index method yielded highest accuracy of 77.8% by ROC curve analysis, and the specificity, sensitivity, positive and negative predictive values were 95.7%, 46.2%, 85.7%, and 75.9%. In vitro chemosensitive group showed higher response rate (85.7% vs. 24.1%) (P=0.005) compared to chemoresistant group. ATP-CRA could predict clinical response to paclitaxel and cisplatin chemotherapy with high accuracy in advanced gastric cancer patients. Our study supports the use of ATP-CRA in further validation studies.