Two experimental treatments are used to study the effects of auction risk across five mechanisms. The first canonical, baseline treatment features only strategic risk and replicates the standard results that overbidding relative to the risk neutral Nash equilibrium is prevalent in all common auction mechanisms except for the English auction. We do not find evidence that bidders' measured risk preferences can explain these patterns of overbidding. To enhance salience, we introduce a second novel treatment with external risk. This treatment captures the risk, prevalent in online auctions, that winners will not receive a good of value. We find that dynamic auctions-including the English-are particularly susceptible to overbidding in this environment. We note that overbidding is somewhat diminished in later periods and that our results may thus have particular relevance for bidders who are not highly experienced or who have not directly experienced losses. We conclude with a brief discussion of research implications.