In the steel industry, especially alloy steel, creating different defected product can impose a high cost for steel producers. One common defect in producing low carbon steel grades is Pits & Blister defect. To eliminate this drawback, we need to grind the surface of the product. In some cases, the severity of defects may lead to scrap part of the product. Grinding cause waste of time and cost of production will be increased. Incidence of defects is related to several factors including material analysis and production processes. In this study we want to create a model to predict this fault with data mining methods including decision tree, neural network and association rules. We will compare the efficiency and accuracy of these models and select the appropriate model. In this study, methodology used to perform data mining is CRISP (Cross Industry Standard Process for Data Mining). To create a decision tree and neural network respectively, entropy method and 24 hidden nodes are used. And to discover association rules, four members itemset is used. And applying data mining on data received from Iran Alloy Steel Company, the model created using the decision tree, has higher accuracy.