After Freeman and Tse (1992), and Hayn (1995) indicated that positive-and negative-earning companies have different responses toward earning compositions based on the negative-earning companies¡¦ information content, Chamber (1996), Subramanyam and Wild (1996), Lipe, Bryant and Widner (1998) proposed the relationship between negative and future earning. In addition, Sloan (1996) suggested that accruals in the market have been over-valued. Xie (2001), however, pointed out that over-valued situation comes from manipulable earning compositions. Recently, Dopuch (2005), and Joos and Plesko (2002) reexamined the properties of negative-earning samples and found that investors¡¦ attitude on negative earnings will depend on predicted reversal possibility for valuation of a company. However, Taiwan has no studies on accrual evaluations after differentiating positive and negative earning for companies. This study adopts Mishkin (1983) evaluation model on abnormal accruals to analyze Taiwan 1391 listed company for the sample period from 1986 to 2005. The empirical results suggest that ERCs¡¦ of negative-earning companies are not significant; if the companies are differentiated by persist negative earning and transitory negative earning, investors tend to overprice transitory negative earning. In addition, investors tend to overprice the (discretionary) accruals positive-earning companies, but make accurate evaluation toward persistent earning companies with low reversal possibility. This study also found that persistency of positive-earning companies is longer than negative-earnings. Finally, by using logistic regression to differentiate the possibility of negative-earning accruals, there is no significant difference between the logistic regression model and cash flow classification from the empirical test. Based on the findings, how Taiwan security market response and evaluate discretionary accruals can be further understood.