Although the subject of power losses in distribution systems is as old as electric power industry history, but its importance increases everyday. Economic, technical and environmental effects of power losses, especially in deregulated systems is indispensable. The first step of each action in respect to power losses is measurment, calculation or estimation of power losses. According to practical difficulties of accurately measuring of power losses in distribution systems, major electrical utilites have turned to the use of methods for estimating power losses. The accuracy of different methods for estimating power losses in the distribution systems are different and therefore have different applications. In this thesis, a Hybrid method of accurately calculating and estimating is used to obtain Power losses in distribution feeders. Therefore, with accurate modeling of components of distribution feeders and employing the three-phase power flow, power losses of test feeders are calculated accurately. Using the obtained patterns from these feeders, the power losses model for distribution feeders are determined based on neural network. Power losses model derived in this thesis is a relatively perfect model that considers all of the important factors including feeder load, feeder length, transformers capacity of feeder, feeder power factor and feeder unbalancy in power losses. Using this model, the power losses of other distribution feeders are estimated with good accuracy. The test feeders IEEE 13, 34, 37, and 123 uses of are used for training and testing of the power lossess model. Keywords: Estimation Losses, Artifical Neural Network, Power Losses Model, Important Factors.