Many financial crisis cases related to the public companies have increased recently, but many investors and creditors are difficult to foresee the financial crisis, especially in the cases with earnings management. In literature, many studies related to earnings management only focus on identifying some related factors which can significantly affect earnings management. Therefore, we can only figure out the correlation between these factors and earnings management. In order to decrease the financial crisis risks derived from earnings management and help the investors avoid suffering a great loss in the stock market, we developed a neural network model to predict the level of earnings management. This study aims to investigate the accuracy of earning management forecast by neural network and decision making tree as well as comparing that by linear models. To these end nine effective variables on earnings management were used as independent variables and discretionary accruals as dependent variables. From four industries: agriculture, pharmaceutical, textile and petroleum, 36 firms selected during 2006 to 2013. The least squares regression for linear model, generalized feed forward neural network and decision making tree c5.0, cart were applied for data mining. The results indicated that neural network and decision making tree has the least error in forecasting earnings management than more accurate linear methods. Concerning the relationship between dependent and independent variables, it is said that earning managements by discretionary accrual variables of the prior period (DAI), non-discretionary accruals of prior period or threshold performance (THOD) and risk (Risk) in four linear models, neural network, C5.0 trees and cart has the most correlation.