This study examines the market valuation of accounting earnings during the period before it is publicly revealed that the earnings are fraudulent. Using both cross‐sectional and time‐series valuation models, we first find that the market accords less weight to earnings when the accounting numbers are fraudulent. We also show that the market better anticipates the presence of fraud when there is information in the public domain indicating a high ex‐ante risk of fraud. Our findings suggest that investors are able to accurately assess the probability of fraud and that such assessments affect the market's valuation of earnings even before it is publicly announced that fraud has occurred.