The invention relates to a prediction method of benzo [a] pyrene release amount in main stream smoke of flue-cured tobacco leaves. The method comprises the steps of subjecting tobacco leaves to be cured firstly to sample pre-processing such as manual piece tearing, stalk removing and tobacco cutting; detecting 7 chemical components (potassium, palmitic acid, malonic acid, belladonna, stearic acid, malic acid and chlorogenic acid) of a sample to be detected; calculating network values of 7 nodes of a hidden layer according to the detection results of the 7 chemical components and coefficients of input layers of a model; converting the network values of the 7 nodes of the hidden layer to output values of the 7 nodes of the hidden layer; and calculating to obtain a predicted value of the benzo [a] pyrene release amount in the smoke according to the output values of the 7 nodes of the hidden layer and coefficients of output layers of the model. According to the method, the possible benzo [a] pyrene accumulated content in cigarette finished products produced in future can be predicted effectively according to raw materials of the flue-cured tobacco leaves, so that raw material choosing in a producing process is guided.