We sought to identify a serum miRNA expression profile to improve disease surveillance and to predict post-operative disease recurrence for stage II/III colorectal cancer (CRC) patients. Using the TaqMan Low-Density Array (TLDA), we performed an initial survey to analyze 749 miRNAs in the pooled serum of 20 paired pre- and post-operative CRC patients and 20 matched normal subjects. Using individual RT-qPCR verification in 175 stage II/III CRC patients, we identified that miR-145, miR-106a and miR-17-3p were significantly differentially expressed between pre- and post-operative CRC patients and between pre-operative CRC patients and normal controls (P < 0.0001). The area under the ROC curve (AUC) for the three-miRNA panel was 0.886 (95% CI 0.850-0.921) for discriminating between pre-operative CRC patients and normal subjects and 0.850 (95% CI 0.809-0.891) for discriminating between pre- and post-operative CRC patients. Furthermore, using the Kaplan-Meier method and Cox proportional hazards analysis, we found that miR-17-3p and miR-106a were powerful and independent prognostic indicators and that high levels of these miRNAs were associated with shorter disease-free survival (DFS) (P < 0.0001 for miR-17-3p and P = 0.001 for miR-106a). The present study reveals novel serum-miRNA-based biomarkers for monitoring tumor dynamics as well as for predicting disease recurrence in patients with stage II/III CRC.