The number of automobile recalls in the U.S. has sharply increased in the last decade and a half, and the number of units involved in these recalls are often counted in the millions. In 2004 alone, over 30 million vehicles were recalled in the United States. However, there is no quantitative evidence of the effect of recalls on safety. Without that evidence, the government and insurance companies have been reluctant to request and use more detailed recall information to increase correction rates. In this paper we empirically quantify the effect of vehicle recalls on safety using repeated cross-sections on accidents of individual drivers and aggregate vehicle recall data, to construct synthetic panel data on individual drivers of a particular vehicle model. We estimate the effect of recalls on the number of accidents, and find that recalls of a particular model reduce accidents of that model by 20%. Recalls classified as "hazardous," and those initiated by foreign manufacturers are more effective in reducing accidents. We also find that vehicle models with recalls with higher correction rates have on average less accidents in the years following a recall, which indicates the importance of the role of drivers' behavior regarding recalls, on safety. The latter suggests that society as a whole, individual drivers, and insurance companies, could benefit from an initiative to take into account recall correction behavior when pricing auto insurance.