A medical data analysis method and device. The method comprises: receiving a physical sign parameter of an object; locating, by using the physical sign parameter as a characteristic, a subspace in which the object is located in a medical data set; analyzing the probability p1 that the object is located in its own subspace by analyzing the semantic consistency of the subspace in which the physical sign parameter of the object is located. In addition, the probability P2 that the object is at its own node may also be analyzed on the basis of the evidence delivery score of the physical sign parameter of the object on a medical knowledge map. The probability P that the object is in/at its own subspace or node may also be determined on the basis of the probability P1 and the probability P2: P=α×R1+(1-α)×R2, wherein α is an harmonization parameter, and 0<α<1. The solution can improve the analysis accuracy, and can also analyze a patient's condition according to symptoms quickly, so as to guide the patient to targeted examinations, thereby reducing costs and improving efficiency.