Although static characterization of reservoirs is an inevitable part of any reservoir studies, the most robust models of the reservoirs can be obtained through integrating static and dynamic data. The following study which is done in a heterogeneous carbonate reservoir utilizes the capillary pressure and relative permeability data to verify the task of static rock typing and investigate the role of hydraulic units in capillary pressure and relative permeability modeling. For this purpose, at first, various rock typing techniques are applied to the field data to seek the best method which has the most consistency with capillary pressure curves. Using Desouky method which is based on hydraulic unit concept, Leverett J-function was then correlated with the normalized water saturation for each rock type to normalize and estimate capillary pressure curves. An excellent agreement exists between the measured data and calculated ones approving the applicability of Desouky method in heterogeneous carbonate reservoirs. Then, using porosity, permeability, wetting phase saturation and some functional links as input data, artificial neural networks (ANN) and specifically the back propagation feed forward neural network is applied as a new approach to predict relative permeability. However, instead of constructing the model for the entire depth of the well, discrete models are developed for each hydraulic unit individually. This idea is generated from the fact that each hydraulic unit has its own set of relative permeability curves and each set differs from one unit to another so the task of separating the models can reduce the discrepancies of the data and thus providing the more accurate synthetic models for predicting relative permeability. Results obtained from this model are compared with those of applying multiple regressions and also ANN method for the entire depth of the well. This comparison showed that the new approach introduced in this study can enhance relative permeability prediction considerably.