This study begins with the analysis of the assumptions of the Linear Earnings-Returns Model, and points out the inherent disadvantages of the linear model. So we assume that the relation of unexpected earnings and unexpected stock returns is nonlinear. Nonlinear Earnings-Returns Model can provide significantly higher explanatory power than linear model. According to the natures of unexpected earnings and unexpected stock returns, we employ inverse tangent (arctan function) as the basic model of our empirical study. The related tests show that nonlinear model has more significant goodness-of-fit than linear model. After excluding the transient items from earnings, the goodness-of-fit of nonlinear model drops a lot. This proves that the transient items may be one of the reasons that cause the nonlinear relation of Earnings-Returns.