Update publication information. Early diagnosis is crucial to improve outcomes for pancreatic cancer (PC) patients, however the lack of specific and validated biomarkers for the disease remains challenging. In this study, we analyzed the serum proteome of 12 PC patients and 76 healthy controls (HC), using magnetic bead-based weak cation exchange and matrix-assisted laser desorption ionization-time of flight mass spectrometry. Next, we established supervised neural network (SNN) algorithm model by ClinProTools and identified the candidate biomarker using liquid chromatography-electrospray ionization-tandem mass spectrometry. Finally, candidate biomarker was validated in tissue samples. The SNN algorithm model discriminated PC from HC with 92.97% sensitivity and 94.55% specificity. We identified 115 differentially expressed peptides, nine of which were significantly different (P<0.005). We identified three upregulated peaks (peaks 2, 8 and 9) and six downregulated peaks (peaks 1, 3, 4, 5, 6 and 7) in samples from PC patients, compared with HC samples. Interestingly, one peptide demonstrated the most significant changes between PC and HC samples, which we identified as regions of RNA-binding motif protein 6 (RBM6). In subsequent tissue analysis, we verified RBM6 expression was significantly higher in PC tissues than paracancerous tissue. Our results indicate that RBM6 might serve as a candidate diagnostic biomarker for PC.