This paper studies the revisions of the expenditure approach variables of the National Accounts, the most popular tool in macroeconomic forecasting and analysis. Preliminary figures get most attention. But how reliable are they? Generally, preliminary figures underestimate growth rates. Some revision distributions are significantly non-normal, being leptokurtic, which translates to frequent outliers. The dispersion of the revisions of some variables is so large that the general usefulness of these variables can seriously be questioned. For many variables, revisions are correlated with the business cycle, i.e. growth rates are revised upwards in upturns and downwards in downturns. Plausible reasons are: biased imputation rules, prudent judgment and a sampling problem. Opposite signs and indications of acceleration/deceleration of preliminary and final figures are recorded and discussed. Different seasonal patterns further emphasize the incongruence between preliminary and final data. This may have serious consequences when analyzing seasonally adjusted data. Consistency of vintages would require cointegration and a common stochastic trend, but none was found on any frequency. A “final” figure may be substantially revised years later. The “latest” long-range growth is different from final and preliminary growth rates. These later revisions are primarily due to new classification rules. An international comparison completes the study. Positive bias appears in many countries’ GDP revisions. Correlation with the business cycle is common among European countries. In the conclusions, some proposals are given on how to improve the usefulness of preliminary data.