The Marshall Stability of Asphalt concrete is one of the most important parameters in mix design and quality control. This property depends on many factors such as gradation, percentage of crushed aggregates, Asphalt content and construction quality. In this research, the variation of Marshall Stability with Asphalt content is simulated using Artificial Neural Networks (ANNs) with a Levenberg- Marquardt Back Propagation (LMBP) training algorithm. The percentage of crushed aggregates; the percentage passing through sieve numbers 200, 50, 30, 8, 4 and 1/2 inch, and the percentage of Asphalt content are considered as network inputs and Marshall Stability as the network output. In the first stage, the maximum generalization ability of each network with a specified number of neurons in the hidden layer is determined. Comparing these maximum values reveals that the network with 8 neurons in the hidden layer has the maximum generalization ability. In the second stage, the variation of Marshall Stability with Asphalt content is simulated by applying a sensitivity analysis to the network with the maximum generalization ability. This simulation is in good agreement with theory.