SFAS 142 requires firms to use fair-value estimates to determine goodwill impairments. Watts (2003) and Ramanna (2007) argue the unverifiable nature of those fair-value estimates gives firms discretion to manage impairments. We test this argument in a sample of firms with market indications of impairment (firms with book goodwill and market-to-book ratio below one). We find that the frequency of non-impairment in this sample is about 71%, and that non-impairment is increasing in financial characteristics predicted to be associated with greater unverifiable fair-value-based discretion. To investigate whether non-impairment is associated with managers producing on average better estimates of goodwill than the market, we test whether non-impairment increases in industries with higher average information asymmetries. We fail to find evidence consistent with this proposition.