In present study, three roughness parameters defined in the Abbott-Firestone or bearing area curve, Rk, Rpk and Rvk, were modeled for the rough honing processes by means of Artificial Neural Networks (ANN). Input variables were grain size and density of abrasive, pressure of abrasive stones on the workpiece's surface, tangential or rotation speed of the workpiece and linear speed of the honing head. A back propagation algorithm was used for training the networks. Two strategies were considered, use of either one network for modeling the three parameters at the same time and use of three networks, one for each parameter. Best network was chosen among different structures, having either one or two hidden layers. When one network is considered, best solution corresponds to two hidden layers having 26 and 11 neurons. However, overall best neural network consists of three networks, one for each roughness parameter, with one hidden layer having 25, 9 and 5 neurons respectively.