Collapse, defined as the additional deformation of compacted soils when wetted, is believed to be responsible for damage to buildings resting on compacted fills, as well as failure in embankments and earth dams. In this paper, three different types of neural networks, namely, conventional Back-Propagation Neural Network (BPNN), Recurrent Neural Network (RNN) and Generalized Regression Neural Network (GRNN) are employed as computational tools to predict the amount of collapse and to investigate the influence of various parameters on the collapse potential. To arrive at this goal, 192 series of a single oedometer test were carried out on three soils with different initial conditions and inundated at different applied pressures. The test results were used to prepare the necessary database for training the neural network. Similar test results available in literature were also included in the database to arrive at a total of 330 sets of data. A comparison of the network prediction for collapse potential with some available models shows the superiority of the network in terms of the accuracy of prediction. Moreover, by analyzing the network connection weights, the relative importance of different parameters on collapse potential was assessed. Based on this analysis, for a given soil type, the initial dry unit weight, \gamma_d, is the most important factor influencing collapse potential.