Evapotranspiration is one of the parameters affecting the watershed water balance and as a basic parameter in hydrologic cycle.Being a function of different metrological parameters and their interactions, evapotranspiration is a complex, nonlinear phenomenon. Preprocessing input parameters to select appropriate combinations is complex when modeling nonlinear systems. Therefore, in this study artificial neural networks MLP, FF and MNN for evapotranspiration modeling and in this context of MATLAB software was used. By using the climatic data in Ahvaz climateyears 1988-2014, the average values of reference crop evapotranspiration was calculated by FAO Penman Montith standard.Then these values was used as output targets of different networks with conventional structures defined and trained. Eventually capability of estimation of evapotranspiration network by using the part of the data network is not used in the design or training, was evaluated. With reviews was conductedshowed who using only the daily average temperature as an input parametercan be the reference evapotranspiration using three types accurately estimate network. Also compares the results of three network tests showed that FF and MLP networks to MNN for determining reference evapotranspiration are higher accuracy.