During the recent housing recession and financial crisis, mortgage modification has been heavily promoted by the U.S. government as a way to stabilize the housing and the national banking systems. Numerous programs, such as the Home Owners Preserving Equity (HOPE), Home Affordability Modification Program (HAMP), and Home Affordability Refinance Program (HARP), were introduced or enhanced to allow more aggressive modifications than traditionally observed prior to the crisis. Loan modification is believed to be a way to avoid foreclosure and to help borrowers keep their homes. However, the effectiveness of loan modification in preventing eventual foreclosure has not been quantified. In this paper, we use Federal Housing Administration (FHA) modified loans to analyze their re-default risk. We use loan-level data to trace the performance of loans with heavy modifications. We have three major empirical findings. First, the empirical model shows that modified loans tend to have much higher re-default risk than otherwise identical never-defaulted loans. Second, the re-default model shows that re-default hazard is less sensitive to traditional risk drivers, compared with non-modified loans. Third, the re-default risk declines initially with the magnitude of the payment reduction associated with the modification received. However, as the payment reduction becomes substantial, the probability of re-default increases. Our empirical results suggest payment reduction is most effective around the 10% to 30% level, in order to reduce re-default risk. The effect is relatively flat between the 30% to 40% level. Payment reduction beyond the 40% level increases re-default risk, controlling for all observable variables. These findings have profound implications in how lenders should design optimal modification policies.