The Libor is a vital benchmark interest rate formed by an unregulated survey of sixteen large banks. These banks own many Libor-referencing contracts and therefore have strong incentives to submit misleading quotes to the survey. In this paper, I measure how much they distorted the Libor between 2005 and 2009. To do so, I specify a model of strategic manipulation which captures the key features of the Libor aggregation mechanism. I use a two-step estimator to back out banks' unobserved portfolio exposures through their revealed preferences. I find the Libor was accurate prior to the 2007 financial crisis, but was subsequently distorted downward by eight basis points, which cost U.S. municipalities $455 million over the sample period. I propose an alternative aggregation mechanism and find it removes virtually all the systematic bias due to manipulation.