The Kleihauer-Betke (KB) test is the standard method for quantitating fetal-maternal hemorrhage in maternal care. In hospitals, the KB test is performed by a certified technologist to count a minimum of 2000 fetal and maternal red blood cells (RBCs) on a blood smear. Manual counting suffers from inherent inconsistency and unreliability. This paper describes a system for automated counting and distinguishing fetal and maternal RBCs on clinical KB slides. A custom-adapted hardware platform is used for KB slide scanning and image capturing. Spatial-color pixel classification with spectral clustering is proposed to separate overlapping cells. Optimal clustering number and total cell number are obtained through maximizing cluster validity index. To accurately identify fetal RBCs from maternal RBCs, multiple features including cell size, roundness, gradient, and saturation difference between cell and whole slide are used in supervised learning to generate feature vectors, to tackle cell color, shape, and contrast variations across clinical KB slides. The results show that the automated system is capable of completing the counting of over 60,000 cells (versus ∼2000 by technologists) within 5 min (versus ∼15 min by technologists). The throughput is improved by approximately 90 times compared to manual reading by technologists. The counting results are highly accurate and correlate strongly with those from benchmarking flow cytometry measurement.