The investigation on a three-span steel plate composite continuous girder bridge was conducted to obtain the strain time response under the moving load of a single heavy vehicle, in order to verify the effectiveness of strain influence line for model modification. An objective function was constructed by measured and calculated strain influence lines. The back propagation (BP) neural network was constructed to do self-training, by taking the micro-strain test value at the control point of the influence line form as the input layer parameter, and the structural geometry information and material characteristic value of finite element model as the output layer parameters. The parameters to be modified were predicted and the bridge model modification was carried out, on the basis of the self-trained BP neural network. Results showed that the proposed finite element model modification method can reduce the modelling error caused by the uncertainty of the real structure, the revised optimization model was closer to the real structure than the initial model, and the relative error of the objective function was reduced by 29%. The model parameter modification method based on BP neural network can be used to predict the parameters of the finite element model.