We studied on possible association between the tumor marker (CEA, CA 19-9, and SPan-1) change and the clinical outcome after treatment with gemcitabine (GEM) in 23 patients with unresectable or recurrent pancreatic cancer. GEM was administered intravenously at a standard dose of 1000 mg/m2 weekly. One course consisted of weekly administration for 3 weeks followed by 1 week's rest. When the adverse effect did not allow the weekly administration, GEM was given bi-weekly without dose modification. Objective responses were evaluated by computed tomography and tumor marker change. Two or more courses were given for only 6 (26.1%) patients. The number of patients, administered GEM 6 or more times including by the weekly and bi-weekly method, was 12 (52.2%). Antitumor effects were evaluable in 16 patients. The clinical efficacies were 1 partial response (PR), 6 stable disease (SD), and 9 progressive disease (PD). Decreases in tumor marker levels were recognized in 9 of the 16 patients. The median survival time (MST) of the PR+NC group was significantly longer than that of the PD group (9 vs 3.5 months; p=0.0151). MST of those in the decreasing tumor marker group was significantly longer than the group with no decreases in the tumor markers (7.0 vs 5.5 months; p = 0.0478). The adverse effects at grade 3 or more were 4 (17.3%) leukopenia, 2 (8.7%) thrombocytopenia, and 1 (4.3%) skin toxicity. In conclusion, the tumor marker change after GEM treatment may be a predictor of preferable prognosis in patients with pancreatic cancer.