Breast cancer is a highly malignant disease in women. A convenient screening tool with high accuracy for early detection, not only in high-risk individuals but in the general population, is necessary. Two hundred breast cancer patients, 100 healthy controls and 100 hyperplasia patients were enrolled in this study. Samples were randomly assigned into training or testing cohorts. The receiver operating characteristic curve was used to explore the optimal concentration of cell-free DNA (GAPDH) in the training cohort and the cut-off point was validated in the testing cohort. The results showed that both in the training and testing cohorts, the overall accuracy of classification between cancer, healthy controls and hyperplasia was higher than 0.9. The sensitivity, specificity, positive predictive value and negative predictive value also reached 0.9, with the exception of the negative predictive value in the testing cohort. This study provides useful information on the use of concentration of free DNA for breast cancer detection. These findings need to be validated in a large prospective trial prior to clinical application.