This paper proposes a new application of neural networks in telecommunications network simulation. A high-level abstracted analytical model, based on intensive investigation of packet queuing behavior, substantially speeds up the basic simulation. Comparing the results from the model against the behavior of a testbed leads to some difference between the model results and the experimental validation, an expected result given the level of abstraction. A neural network is applied to learn the relation between the model parameters and the output difference, and neural network prediction is used to 'fine-tune' the model accordingly. Results indicate that the proposed hybrid method (using the neural network to tune the abstracted model) achieves fast simulation and remains accurate. This approach is particularly useful in the area of large-scale network designing and planning, where concern is more about the overall performance of the network than the detailed structure of a network node.