Long non-coding RNAs (lncRNAs) have been demonstrated to be aberrantly expressed in several types of tumor, and dysregulated lncRNAs are suggested to play a prognostic role in breast cancer (BC). Estrogen receptor (ER) status is a prognostic factor in patients with ER-negative BC, which is associated with poor prognosis. Thus, the present study developed a prognostic lncRNA signature specifically for ER-negative BC, in order to predict the risk of post-surgery relapse and improve patient prognosis. A gene expression profile containing 1,631 lncRNAs was obtained by investigating and integrating publicly available cohorts of BC. Subsequently, a nine-lncRNA signature was developed and validated in two independent cohorts via the Cox regression model. Using the nine-lncRNA signature, patients in the discovery cohort were divided into high-and low-risk groups, with significantly different disease-free survival [DFS; hazard ratio (HR)=2.718, 95% confidence interval (CI)=2.115-3.494, P<0.0001]. Receiver operating characteristic curve analyses demonstrated that the area under the curve reached 0.908. Similar results were obtained in the two independent cohorts (HR=1.499, 95% CI=0.950-2.365, P=0.04; HR=1.262, 95% CI=1.056-1.510, P=0.01), respectively. Furthermore, the nine lncRNAs were demonstrated to play important roles in the cell invasion and metastasis of different types of tumor. The differentially expressed genes (DEGs) identified between the high-and low-risk groups were consistently high in the discovery and validation cohorts. Functional analysis indicated that these DEGs, as well as genes co-expressed with the nine lncRNAs, were involved in cancer-associated signaling pathways, all of which provide further evidence for the predictive ability of the nine-lncRNA signature. Overall, the present study developed a novel prognostic biomarker for ER-negative BC.