Novel, non-invasive biomarkers to diagnose breast cancer with high sensitivity and specificity are greatly desired. Circulating microRNAs (miRNAs) show potential for breast cancer detection, but the existing results appear to be mixed. Using microscale serum, we established a novel serum-direct multiplex detection assay based on RT-PCR (SdM-RT-PCR). Ninety-three miRNAs dysregulated or with functions in breast cancer were selected as candidates, and additional 3 miRNAs were chosen as endogenous controls. We first conducted miRNA profiling of these 96 miRNAs by SdM-RT-PCR using the sera of 25 breast cancer patients at diagnosis prior to treatment and 20 age-matched healthy controls. miRNAs showing significantly different expression levels between patients and controls were further analyzed using a logistic regression model. A miRNA signature was validated in an independent set of 128 serum samples composed of 76 breast cancer patients and 52 healthy controls. In the discovery stage, we identified 23 miRNAs as significantly dysregulated in breast cancer patients compared with healthy controls. Of these, 10 miRNAs were previously identified as dysregulated in breast cancer; 14 miRNAs remained significant after P-values were adjusted by both correction methods. Principal component analysis and hierarchical clustering of these miRNAs separated patients from controls. Furthermore, the 3-miRNA signature (miR-199a, miR-29c, and miR-424) with the highest diagnostic accuracy for distinguishing breast cancer patients from controls by ROC curve analysis (AUC = 0.888) was successfully confirmed in the validation set (AUC = 0.901). Our data demonstrate that the SdM-RT-PCR assay is an effective breast cancer profiling method that utilizes very small volumes and is compatible with Biobank. Furthermore, the identified 3-miRNA signature is a promising circulating biomarker for breast cancer diagnosis.