Trabectedin was recently approved for patients failing doxorubicin, the standard treatment for advanced/metastatic sarcoma. This retrospective study aimed to compare trabectedin efficacy between compassionate use in unselected patients and clinical trials. From May 1999 to January 2006, 92 patients were treated at the Centre Léon Bérard, either in phase II studies or on a named patient compassionate basis. All cases were retrospectively analyzed to assess trabectedin efficacy in terms of response, progression-free, and overall survival.The objective response rate was 10% (N=9): 4% (N=2) for patients treated in compassionate use program and 16% (N=7) for those in clinical trials (P=0.18); 26 (28%) patients had stable disease for at least 6 months, 11 (23%) in the compassionate group and 15 (33%) in clinical trials. Median progression-free and overall survivals were, respectively, 2.2 [95% confidence interval (CI): 1.9-3.6] and 8.9 (95% CI: 6.4-14.2) months for all patients, 2.3 (95% CI: 1.9-4.3) and 10.4 (95% CI: 6.9-24.2) months for patients in clinical trials and 1.8 (95% CI: 1.4-3.4) and 6.4 (95% CI: 3.3-14.2) months for patients under compassionate treatment. In this retrospective analysis, the reported grade 3-4 toxicities were increased transaminase (34 patients, 37%) and neutropenia (38 patients; 42%). Higher efficacy was observed in phase II studies than with compassionate treatment, but no significant difference remained after adjustment in multivariate analysis for performance status, a well-established prognosis factor. The safety and tolerability of trabectedin shown in clinical trials is confirmed for patients in real-life situation treated in compassionate use programs, but its benefit is higher for patients with performance status 0-1.