Revisions of budget balances in Europe could be particularly worrisome as adherence to multilateral surveillance rules is judged upon initial data releases. We use a pool of real‐time vintages of data for 15 EU countries over the period 1995–2008. Our main findings are: (i) preliminary releases are biased and nonefficient predictors of subsequent releases, (ii) such systematic bias in revisions is a general feature of the sample, (iii) Eurostat's decisions explain a significant share of the bias and provide some evidence of window dressing practices, and (iv) expected real gross domestic product growth, political cycles, and the strength of fiscal rules also contribute to explain revision patterns.