Metastasis-related mRNAs have showed great promise as prognostic biomarkers in various types of cancers. Therefore, we attempted to develop a metastasis-associated gene signature to enhance prognostic prediction of breast cancer (BC) based on gene expression profiling. We firstly screened and identified 56 differentially expressed mRNAs by analysing BC tumour tissues with and without metastasis in the discovery cohort (GSE102484, n = 683). We then found 26 of these differentially expressed genes were associated with metastasis-free survival (MFS) in the training set (GSE20685, n = 319). A metastasis-associated gene signature built using a LASSO Cox regression model, which consisted of four mRNAs, can classify patients into high- and low-risk groups in the training cohort. Patients with high-risk scores in the training cohort had shorter MFS (hazard ratio [HR] 3.89, 95% CI 2.53-5.98; P < 0.001), disease-free survival (DFS) (HR 4.69, 2.93-7.50; P < 0.001) and overall survival (HR 4.06, 2.56-6.45; P < 0.001) than patients with low-risk scores. The prognostic accuracy of mRNAs signature was validated in the two independent validation cohorts (GSE21653, n = 248; GSE31448, n = 246). We then developed a nomogram based on the mRNAs signature and clinical-related risk factors (T stage and N stage) that predicted an individual's risk of disease, which can be assessed by calibration curves. Our study demonstrated that this 4-mRNA signature might be a reliable and useful prognostic tool for DFS evaluation and will facilitate tailored therapy for BC patients at different risk of disease.