Abstarct: According to there are 90% petroleum reserves in carbonate rocks, textural identification of carbonate rocks is one of most important in petroleum exploration in carbonate rocks. In traditional methods, since sampling in determined depth of the well, thin sections were prepared. Then petrology specialist study on thin section and nominate them. This procedure is very Time-consuming and costly and because of it is Time-consuming and difficult; probability of human error is very high. Application of a method baxsed on computer and artificial intelligence can improve problems due time and cost. In the study, software baxsed on neural network and image processing is created. The software get digitized microscopic images from surface of thin section with 25x magnitude as input and determine type of carbonate rock. 400 thin sections were used to neural network training and each 100 thin sections belong to one of mudstone, wackstone, packstone and granstone groups. To validate neural network and software results 400 thin sections were used and these thin sections different with training thin sections. According to validation results accuracy of software to identification of each one of mudstone, wackstone, packstone and grainstone groups is 100, 76, 71 and 87 respectively.