Objective: The aim of this study was to evaluate patient and tumor characteristics in 102 patients with unknown primary tumors (UPT) prospectively treated with a combination of carboplatin, doxorubicin, and etoposide, to identify clinical variables predictive of response and survival. Patients and Methods: The association between clinical characteristics and outcome was evaluated by univariate and multivariate analysis: χ2 test and logistic regression analysis were used to study variables predictive of response, and survival analysis, comparison of survival curves and Cox multiple regression analysis to study variables predictive of survival. Results: We obtained 26.5% objective responses (95% confidence interval: 18.2–36.1%) and a median survival of 9 months (95% confidence interval: 7–11 months). Several variables were associated with response to treatment and survival at univariate analysis. At multivariate analysis the number of tumor sites, bone/visceral involvement and epithelial tumor markers were significantly predictive of response; presence of pain, serum alkaline phosphatase, carboplatin AUC and response to treatment were significantly associated with survival. Conclusions: The identification of variables that can predict prognosis and response to treatment in patients with UPT may be useful to offer aggressive treatment to sensitive subsets of patients and provide therapeutic alternatives to those with a low probability of benefiting from standard treatment. In our patients the use of carboplatin AUC higher than 6 and response to treatment were the most important factors associated with prognosis, together with presence of pain and serum alkaline phosphatase. However, larger series and identification of new disease markers are necessary to better define predictive and prognostic variables in UPT patients.