The size distribution of firms in manufacturing industries has long been a matter of interest in industrial organization. Distribution in different industries show considerable regularity that static economic theory fails to explain. Stochastic growth models appear to provide some insights, but empirical tests of the log-normal or Pareto distributions have been inconclusive. This paper draws on market share data for over three hundred U.S. manufacturing industries and analyses the distribution of largest firm sizes. A statistical test of the Pareto hypothesis, rather different from previous tests in the literature, decisively rejects that hypothesis as a general explanation for the upper tail of the distribution. Instead, great diversity among distributions is found, and the regularities that do emerge imply a greater clustering of large firms than predicted by theory.