Quantifications of gains from trade in heterogeneous firm models assume that productivity is Pareto distributed. Replacing this assumption with log-normal heterogeneity retains some useful Pareto features, while providing a substantially better fit to sales distributions-especially in the left tail. The cost of log-normal is that gains from trade depend on the method of calibrating the fixed cost and productivity distribution parameters. When set to match the size distribution of firm sales in a given market, the log-normal assumption delivers gains from trade in a symmetric two-country model that can be twice as large as under the Pareto assumption.