To identify the best early warning index for world grain markets, this study compares the performance of early warning indices computed from a regression model, probit model and signal approach model. The main result of this study is that, in the view of predictive power, the early warning indices from the regression model or probit model show better performance than that from the signal approach model, which most of early warning systems constructed in Korea have adopted. In addition to this finding, this study suggests using principal component analysis as a way to sum up information of many explanatory variables into several variables and develops a way to guarantee early warning indices obtained from co-moving variables by using predictive values of co-moving variables.