Identifying the types of body fluids left at the crime scene can be essential to reconstructing the crime scene and inferring criminal behavior. MicroRNA (miRNA) molecule extracted from the trace of body fluids is one of the most promising biomarkers for the identification due to its high expression, extreme stability and tissue specificity. However, the detection of miRNA markers is not the answer to a yes–no question but the probability of an assumption. Therefore, it is a crucial task to develop complicated methods combining multi-miRNAs as well as computational algorithms to achieve the goal. In this study, we systematically analyzed the expression of 10 most probable body fluid-specific miRNA markers (miR-451a, miR-205-5p, miR-203a-3p, miR-214-3p, miR-144-3p, miR-144-5p, miR-654-5p, miR-888-5p, miR-891a-5p and miR-124-3p) in 605 body fluids-related samples, including peripheral blood, menstrual blood, saliva, semen and vaginal secretion. We introduced the kernel density estimation (KDE) method and six well-established methods to classify the body fluids in order to find the most optimal combinations of miRNA markers as well as the corresponding classifying method. The results show that the combination of miR-451a, miR-891a-5p, miR-144-5p and miR-203a-3p together with KDE can achieve the most accurate and robust performance according to the cross-validation, independent tests and random perturbation tests. This systematic analysis suggests a reference scheme for the identification of body fluids in an accurate and stable manner.