Pancytopenia is a common blood disorder defined as the decrease of red blood cells, white blood cells and platelets in the peripheral blood. Its genesis mechanism is typically complex and a variety of diseases have been found to be capable of causing pancytopenia, some of which are featured by their high mortality rates. Early judgement on the cause of pancytopenia can benefit timely and appropriate treatment to improve patient survival significantly. In this study, a serum surface-enhanced Raman spectroscopy (SERS) method was explored for the early differential diagnosis of three pancytopenia related diseases, i.e., aplastic anemia (AA), myelodysplastic syndrome (MDS) and spontaneous remission of pancytopenia (SRP), in which the patients with those pancytopenia related diseases at initial stage exhibited same pancytopenia symptom but cannot be conclusively diagnosed through conventional clinical examinations. The SERS spectral analysis results suggested that certain amino acids, protein substances and nucleic acids are expected to be potential biomarkers for their early differential diagnosis. In addition, a diagnostic model was established based on the joint use of partial least squares analysis and linear discriminant analysis (PLS-LDA), and an overall accuracy of 86.67 % was achieved to differentiate those pancytopenia related diseases, even at the time that confirmed diagnosis cannot be made by routine clinical examinations. Therefore, the proposed method has demonstrated great potential for the early differential diagnosis of pancytopenia related diseases, thus it has significant clinical importance for the timely and rational guidance on subsequent treatment to improve patient survival.