Background: Breast cancer (BRCA) represents the most common malignancy among women worldwide that with high mortality. Radiotherapy is a prevalent therapeutic for BRCA that with heterogeneous effectiveness among patients. Methods: we proposed to develop a gene expression-based signature for BRCA radiotherapy sensitivity prediction. Gene expression profiles of BRCA samples from the Cancer Genome Atlas (TCGA) and International Cancer Genome Consortium (ICGC) were obtained and used as training and independent testing dataset, respectively. Differential expression genes (DEGs) in BRCA tumor samples compared with their paracancerous samples in the training set were identified by using edgeR Bioconductor package followed by dimensionality reduction through autoencoder method and univariate Cox regression analysis to screen genes among DEGs that with significant prognosis significance in patients that were previously treated with radiation. LASSO Cox regression method was applied to screen optimal genes for constructing radiotherapy sensitivity prediction signature. Results: 603 DEGs were obtained in BRCA tumor samples, and seven out of which were retained after univariate cox regression analysis. LASSO Cox regression analysis finally remained six genes based on which the radiotherapy sensitivity prediction model was constructed. The signature was proved to be robust in both training and independent testing sets and an independent marker for BRCA radiotherapy sensitivity prediction. Conclusions: this study should be helpful for BRCA patients’ therapeutics selection and clinical decision.