The previous investigation shows that the combustion behavior of base bleed propellant is classified to three states under the condition of transient depressurization, which are re-ignition,low frequency oscillation combustion and extinguishment, respectively. Variant combustion state was determined mainly by the maximum pressure before transient depressurization (20-90 MPa) and the maximum depressurization velocity (1.2×10³-6×10³ MPa?s-¹) during whole process. The relationship of combustion state between the maximum pressure before transient depressurization and the maximum depressurization velocity is linear separable pattern approximately. Two kinds of perceptron models, single layer and double layers, were constructed according to artificial neural networks. Both of the two peceptron models were trained by using test data. The decision boundary of combustion states of base bleed propellant was obtained, which included both the maximum pressure before transient depressurization and the maximum depressurization velocity factors. The correction and reliably of the peceptron were validated by Monte-Carlo random sampling. The peceptrons could be used to predict the combustion behavior of base bleed