The invention provides a biomass calorific value estimation method based on a BP artificial neural network algorithm. The biomass calorific value estimation method comprises steps that a biomass basic database is read in to acquire data; the neural node quantity of a hidden layer of a neural network is preset; a weight coefficient of an input layer connected with the neural network and the hidden layer is preset; a calorific value of a known biomass is estimated through utilizing the weight coefficient to acquire a prediction result; deviation of the prediction result and a measurement result is calculated; when deviation is smaller than 5%, the present weight coefficient is recorded and stored; a calorific value of an unknown biomass is estimated according to the lastly selected neural node quantity and the weight coefficient. The method is advantaged in that required basic parameters are reduced on the condition that certain precision is guaranteed, estimation errors are further reduced, model coefficients are respectively corrected through employing a reverse error transmission mode, and an estimation result is made to be closer to a measurement result.