This paper formalises the accruals process to model both normal and abnormal accruals under clean surplus accounting. Based on theoretically derived abnormal accruals, I assess the ability of unexpected accruals extracted from the empirical Jones-type models in detecting earnings management. These models include the original Jones (1991) model, the modified Jones model developed in Dechow et al. (1995), and the model in Ball and Shivakumar (2006) that incorporates conditional conservatism. My analysis reveals that unexpected accruals from these models generally do not fully reflect abnormal accruals. I show that a plausible explanation for this could be that abnormal and normal accruals are inherently correlated and they cannot be completely disentangled from each other.