Background: Dysregulated microRNAs (miRNAs) have been identified to be associated with various diseases. However, the relationship between serum miRNAs and the severity of acute pancreatitis (AP) remains unknown. Methods: One hundred twenty-five patients were enrolled and classified into mild AP (MAP, n = 45), moderate severe AP (MSAP, n = 42) and severe AP (SAP, n = 38) groups according to Atlanta 2012. TaqMan low density array (TLDA) technology was initially used in three pooled serum samples from 10 MAP, 10SAP and 10 healthy controls (HCs). The selected miRNAs were subsequently measured individually using quantitative real-time polymerase chain reaction (qRT-PCR) assay. Results: The TLDA identified 395 miRNAs were differentially expressed between the AP patients and the HCs, among which 12 miRNAs were selected for further evaluation. qRT-PCR confirmed that miR-19a, miR-143 and miR-374-5p were significantly upregulated in AP patients (p < 0.001) and increased with disease severity (p < 0.05). Receiver operating characteristic (ROC) curve analysis revealed that three miRNAs could distinguish the SAP patients from the HCs (area under ROC, AUC 0.940-0.943), MAP (AUC 0.754-0.782), and moderate severe AP (MSAP, AUC 0.670-0.686). In addition, multivariate logistic regression revealed that increased serum miR-143 was an independent predictor of developing SAP among MSAP and SAP patients (OR = 6.8, 95% CI 2.0-22.7) and among all AP patients (OR = 4.5, 95% CI 1.8-10.9). Conclusions: These data indicate that an expression signature of three serum miRNAs holds potential as a novel biomarker for the early prediction of SAP.