This paper investigates two competing hypotheses for the accrual anomaly: investment/growth and persistence. Both investment/growth and persistence information in accruals are likely to vary cross-sectionally, depending on a firm's business model, a fact that generates different cross-sectional implications for the accrual anomaly. I find that the magnitude of the accrual anomaly monotonically increases with the investment information contained in accruals, as measured by the co-variation between accruals and employee growth. In industries/firms in which accruals co-vary with employee growth, accruals show strong predictive power for future stock returns. In industries/firms in which accruals show little correlations with employee growth, the accrual anomaly is much weaker. In contrast, the evidence from the cross-sectional analysis is inconsistent with the persistence argument. From the earnings perspective, the evidence on one-year-ahead earnings growth is inconclusive, but the results on longer-term earnings growth support the investment argument but not the persistence argument. Collectively, I conclude that these results support the view that the accrual anomaly is attributable to the fundamental investment information contained in accruals.