Jan Eeckhout (2004) reports that the empirical city size distribution is lognormal, consistent with Gibrat's Law. We show that for the top 0.6 percent of the largest cities, the empirical distribution is dramatically different from the lognormal, and follows a power law. This top part is...