Background: Despite of radical operation, about half of gastric cancer (GC) patients with advanced GC develop peritoneal metastasis (PM) and the patients with PM has poor prognosis. However, because staging laparoscopy was high invasive procedure for the patients, identification of PM by using liquid biopsy can be useful in patients with GC. Methods: We analyzed two genome-wide miRNA expression profiling datasets (GSE164174 and TCGA). We prioritized biomarkers in pretreatment plasma specimens from clinical training and validation cohorts of patients with GC. We developed an integrated exosomal miRNA panel and established a risk-stratification model, which was combined with miRNA panel and currently used tumor markers (CEA, CA19-9, CA125, and CA72-4 levels). Results: Our comprehensive discovery effort identified a 4-miRNA panel that robustly predicted the metastasis, with an excellent accuracy in TCGA dataset (AUC=0.86). We successfully established a circulating exosomal miRNA panel with remarkable diagnostic accuracy in the clinical training (AUC=0.85) and validation (AUC=0.86) cohorts. Moreover, the predictive accuracy of the panel was significantly superior to conventional clinical factors (P<0.01), and the risk-stratification model was dramatically superior to the panel and currently used clinical factors for predicting PM (AUC=0.94, univariate: OR = 77.00, P < 0.01; multivariate: OR = 57.71, P = 0.01). Conclusions: Our novel risk-stratification model for predicting PM has a potential for clinical translation as a liquid biopsy assay in patients with GC. Our findings highlight the potential clinical impact of our model for improved selection and management of patients with this malignancy.