Introduction: SELDI-TOF-MS is a laboratory friendly technique to identify biomarkers for cancer. We explored the application of serum SELDI proteomic patterns to distinguish pancreatic cancer (PC) from chronic pancreatitis (CP), type II diabetes mellitus (DM) or healthy controls (HC). Methods: Sera from 12 HC, 24 DM, 126 PC (84 diabetics) and 61 CP (32 diabetics) were analyzed by SELDI-TOF-MS. Spectra, generated on IMAC-30, were clustered and classified using Chipergen Biomarker Wizard and Biomarker Pattern software. Results: Peaks present in at least 5% of all spectra were selected. Two decision tree classification algorithms, including or not CA19-9 as predictor, were constructed. In the absence of CA19-9 the splitting protein peaks were at 1526, 1211 and 3519 m/z. When CA 19-9 entered the analysis, the former two peaks were maintained as splitters while the 3519 was replaced by CA19-9. The two classification trees performed equally in classifying HC (Se=100%) and DM (Se=100%); CA19-9 tree classified better both CP (Se=89% vs 79%) and PC (Se=63% vs 57%). The specificity of this classification tree was 93%, better of CA 19-9 alone (Se=86%, Sp=65%). We then constructed a classification tree considering only diabetic patients. The optimal tree resulted from the following main splitters: 1211, CA19-9, 7903, 3359, 1802. 100% DM, 97% CP and 77% PC were correctly classified. SELDI-TOF-MS features improved the diagnostic accuracy of CA19-9 (AUC=0.883 for CA19-9; AUC=0.935 for CA19-9 and SELDI-TOF-MS features combined).Conclusion: SELDI-TOF-MS allowed the identification of new peptides which, in addition to CA 19-9, allowed to correctly classify the vast majority of PC patients and to distinguish them from CP or DM.