National accounts data are always revised. Not only recent data, but also figures dating many years back can be revised substantially. This means that there is a danger that an important part of the central bank's information set is flawed for a long period of time. In this paper we present a data base consisting of various vintages of real-time data from 1993Q1 to 2003Q4. We describe the nature of the data revisions, the causes of the revisions, and investigate whether the revisions are true martingale differences, or whether they can be forecasted. In the spirit of Orphanides and van Norden (2002), we analyze how data revisions and model uncertainty affect the reliability of output gap estimates. We also compare Taylor type interest rate rules based on real-time data versus final data and assess the consequences for monetary policy if policy was conducted using this type of interest rate rules. Finally, we analyze the implications of output gap uncertainty for monetary policy using a small New Keynesian macroeconomic model.