We screened an orthotopic nude mouse model of human pancreatic cancer for candidate serum biomarkers and examined their presence in the plasma of pancreatic cancer patients. Nude mice were injected in the pancreas with L3.9pl human pancreatic cancer cells. One week later, the mice were randomized into 4 treatment groups: i) control, saline; ii) oral STI 571; iii) intraperitoneal gemcitabine; and iv) STI 571 and gemcitabine. After 1, 2, and 3 weeks of treatment, sera and tumors were collected from mice in each group as well as uninjected mice. All sera were analyzed by surface enhanced laser desorption ionization mass spectrometry using ProteinChip technology. Protein profiles were analyzed with the Biomarker Wizard software package. The concentration of candidate proteins was evaluated in mouse sera and plasma from 135 pancreatic cancer patients, 7 pancreatitis patients, and 113 healthy volunteers. The combination therapy inhibited tumor growth. A 11.7-kDa protein peak correlating with tumor weight was purified by gel filtration, separated by SDS-PAGE, and identified as mouse serum amyloid A (SAA) by amino acid sequencing and public database searches. The expression of SAA in mouse sera was confirmed by Western blotting and correlated with tumor weight. The level of SAA in plasma of pancreatic cancer patients correlated with clinical stage and was significantly higher than in normal volunteers (mean value: 180.1 microg/ml vs 27.9 microg/ml: P<0.01) or pancreatitis patients. For SAA used as a single tumor marker with a cut-off of 75 microg/ml, the sensitivity for pancreatic cancer was 96.5% and specificity was 31.9%. Our search for specific marker proteins to identify pancreatic cancer was unsuccessful. Although SAA is not specific for pancreatic cancer and not sensitive enough to detect stage I patients, it may be a candidate biomarker for detecting and monitoring the progressive growth of pancreatic cancer.