Triple-negative breast cancers (TNBC) are associated with high risk of early tumor recurrence and poor outcome. Common prognostic biomarkers give very restricted predictive information of tumor recurrences in TNBC. Human serum contains stably expressed microRNAs (miRNAs), which have been discovered to predict prognosis in patients with cancer. The purpose of this study was to identify circulating biomarkers able to predict clinical outcome in TNBC. We performed genome-wide serum miRNA expression and real-time PCR analyses to investigate the ability of miRNAs in predicting tumor relapse in serum samples from 60 primary TNBC. Patients were divided into training and testing cohorts. By Cox regression analysis, we identified a four-miRNA signature (miR-18b, miR-103, miR-107, and miR-652) that predicted tumor relapse and overall survival. This miRNA signature was further validated in an independent cohort of 70 TNBC. A high-risk signature score was developed and significantly associated with tumor recurrence and reduced survival. Multivariate Cox regression models indicated that the risk score based on the four-miRNA signature was an independent prognostic classifier of patients with TNBC. This signature may serve as a minimally invasive predictor of tumor relapse and overall survival for patients with TNBC. This prediction model may ultimately lead to better treatment options for patients with TNBC.