Patients recruited for phase I trials are considered to have a poor prognosis, because the majority of them have already been heavily treated. We examined the survival of lung cancer patients admitted to phase I trials and treated with single new investigational chemotherapeutic agents or new investigational biological modifiers in the Division of Thoracic Oncology of the National Cancer Center Hospital between 1987 and 1993. Eighty-two patients had lung cancer among 121 patients registered in phase I trials. To identify prognostic factors, univariate and multivariate analyses were conducted. Median survival time (MST) from beginning of the phase I trial was 9.4 months, and the response rate was 4.2%. There were 11 (13.4%) early deaths within 3 months, and the death of 1 (1.2%) patient was treatment-related. Univariate analysis demonstrated that performance status, body weight loss, chemotherapy regimen, liver metastasis, number of metastatic sites, prior chemotherapy, and serum levels of hemoglobin, and lactate dehydrogenase were significant prognostic factors for survival. Among these factors, performance status >1, body weight loss >/=10%, and number of the metastatic sites >1 were selected as risk factors in multivariate regression analysis. The low-risk group, which included the 36 patients with no risk factors, had an MST of 13.9 months and an early death rate of 0%. The intermediate-risk group of 31 patients was characterized by patient having only one risk factor. These patients had an MST of 7.6 months and an early death rate of 13%. The high-risk group of 9 patients had two or three risk factors. These patients had an MST of only 1.5 months and a high early death rate of 78%. We conclude that the MST of lung cancer patients who participated in the phase I trials was 9.4 months. Therefore, it appears reasonable to have admitted these patients to the phase I trials. However, as the patients in the high-risk group had a poor outcome and high early death rate, they should not have been admitted to phase I trials. This prognostic model should be validated in other patient series.