To a great extent, engine's runtime of armored vehicle reflects its technical state. By estimating engine's runtime, the remaining service life can be forecasted. In this paper, the virtues of neural network prediction are introduced. Aiming at a certain armored vehicle engine, the BP neural network regression prediction model is constructed based on four typical state parameters including cylinder compression pressures, acceleration time, deceleration time and supply fuel advance angle. The model is trained and validated by samples data. The prediction results indicate that the model is effective and feasible. At last, two prediction problems that need to be studied hard are proposed.