Prediction of crop yield mainly strategic plants such as wheat, corn, rice has always been an interesting research area to agrometeorologist, as it is important in national and international economic programming. The main purpose of such studies is to estimate the crop production a few days or few months before harvesting, using meteorological data. Recently, application of Artificial Neural Networks (ANNs) has been developed as a powerful tool that can compute most complicated equations and numerical analysis to the best approximation. The goal of this study is to apply the ANNs for prediction of dry farming wheat yield. According to the available data and information in different areas of Iran, this research was accomplished using Sararood station data in Kermanshah Province, which has the most complete homogeneous statistic data. The climatic observation data used in this study were mean of daily minimum temperature, extreme of daily minimum temperature, mean of daily maximum temperature, extreme of daily maximum temperature, sum of daily rainfall, number of rainy days, sum of daily sun hours, mean of daily wind speed, extreme of daily wind speed, mean of daily relative humidity, and sum of water requirements that were collected during 1990-1999 for wheat at different phenological stages (11 stages) consisting; sowing, germination, emergence, 3rd leaves, tillering, stem formation, heading, flowering. milk maturity, wax maturity, full maturity. These data were arranged in two forms of matrix: 1- A matrix which its rows are repetition of the statistical years (i) at each phenological stage (j) and its columns are meteorological factors (k). 2- A matrix which its rows are each of the statistical years (i) and its columns are meteorological factors (k) at each phenological stage (j). In fact, statistical years (i), phenological stage (j), and meteorological factors (k) are the basic elements of cubic matrix (Mijk) and the best answer was only derived from the second matrix. Finally, different networks were made for each stage and the optimum values of network parameters were obtained through try and error procedure. It should be noticed that two years of the eight years-data were randomly put away and were not used for network training, in order to test the accuracy of the trained network by comparing actual and estimate data. Abilities of the made model are as follows: 1- prediction of wheat yield with maximum error of 45-60 kg/ha at least two months before full maturity stage (end of stem formation stage). 2- Determination of sensitivity of each phenological stage with respect to meteorological factors. 3- Determination of priority order and importance of each meteorological factors, effective to plant growth and crop yield.