In this paper, we establish a quantitative model based on BP neural network to detect hemoglobin in serum, and put it into practice of hemoglobin measurement, in the result of which various cancer diseases can be completely distinguished. A double-hidden-layer BP neural network model to quantify hemoglobin is established. The correlation coefficient of the prediction set in the model is 0.9838, and its relative deviation is 2.532%. The hemoglobin concentrations in serum samples of breast cancer patients and leukemia patients are obtained by the BP neural network model. A remarkable difference appears in the concentrations between these two sorts of patients (P<0.001), indicating that the model has potential application in distinguishing tumor diseases.