This study presents an intelligent model based on the probabilistic neural networks (PNN) for making a quantitative formulation between seismic attributes and hydraulic flow unit (HFU). Neural networks have been used for several years in reservoir properties estimation. However, their application for hydraulic flow unit estimation from a cube of seismic data is an interesting topic for research. The methodology is illustrated using 3D seismic attributes, petrophysical and core data from 6 wells from the Kangan and Dalan gas reservoirs in the Persian Gulf basin. The methodology introduced in this study is able to estimate HFUs from a large volume of 3D of seismic data. This can increase exploration success rates and reduce costs through the application of the more reliable output results in hydrocarbon exploration programs. Four seismic attributes including acoustic impedance, dominant frequency, amplitude weighted phase and instantaneous phase are considered as the optimal inputs for predicting HFUs from seismic data. The proposed technique is successfully tested in a carbonate sequence of Permian-Triassic rocks from the studied area. The results of this study demonstrate that there is a good agreement between the core and PNN derived flow units. The PNN is successful in modeling flow units from 3D seismic data for which no core data and well logs data are available.