The paper analyzes trade theories and evidence using bilateral, product-level data. It shows that leading trade models fail to explain certain patterns, such as the spatial incidence of "export zeros" (correlated with distance and importer size) and export unit values (positively related to distance). A variant of the Melitz model is proposed that can account for all these facts. In this model, high-quality firms are the most competitive, with heterogeneous quality increasing with firms' heterogeneous costs. The gravity equation relates bilateral trade volumes to distance and country size. However, the many potential explanations for the success of the gravity equation make it a problematic tool for discriminating among trade models. The paper shows that focusing on how the number of traded goods and their prices differ as a function of trade costs and market size is very informative about the ability of trade theory to match trade data. The authors use the most disaggregated data publicly available to establish new facts about U.S. trade. They find that most potential export flows are not present, and the incidence of these "export zeros" is strongly correlated with distance and importing country size. Export unit values are positively related to distance and negatively related to market size. The paper concludes with a variant of the heterogeneous-firms trade model pioneered by Marc Melitz (2003). This model maintains the core structure of Melitz, namely heterogeneity in firms' productivity with fixed market entry costs, but introduces a taste for quality so that the lowest priced goods are not necessarily the most competitive. In this model, firms' competitiveness depends upon their quality-adjusted price and, in equilibrium, higher quality goods are more costly, more profitable and better able to penetrate distant markets. The model's predictions are borne out by the facts established in the data analysis. The paper also shows that the export zero prediction is that the probability that no firm exports in a given code from o to d is increasing in the distance between o and d and in the size of d. The export price prediction is that, within a given code, higher-cost products are less likely to be shipped longer distances and to less remote markets. As a consequence of this composition effect, the average export price within a code will be decreasing in distance, increasing in remoteness, and unrelated to size. The paper also shows that the export zero prediction with linear demand is that the probability that no firm exports in code i from o to d is decreasing in the distance between o and d. Unlike in the EK model, the export price prediction with linear demand has nothing to do with a composition effect. Rather, the prediction is driven by the reduction in markups with distance. The paper also shows that the export zero prediction with linear demand is that the probability that no firm exports in code i from o to d is decreasing in the distance between o and d. The export price prediction is that, within a given code, higher-cost products are less likely to be shipped longer distances and to less remote marketsThe paper analyzes trade theories and evidence using bilateral, product-level data. It shows that leading trade models fail to explain certain patterns, such as the spatial incidence of "export zeros" (correlated with distance and importer size) and export unit values (positively related to distance). A variant of the Melitz model is proposed that can account for all these facts. In this model, high-quality firms are the most competitive, with heterogeneous quality increasing with firms' heterogeneous costs. The gravity equation relates bilateral trade volumes to distance and country size. However, the many potential explanations for the success of the gravity equation make it a problematic tool for discriminating among trade models. The paper shows that focusing on how the number of traded goods and their prices differ as a function of trade costs and market size is very informative about the ability of trade theory to match trade data. The authors use the most disaggregated data publicly available to establish new facts about U.S. trade. They find that most potential export flows are not present, and the incidence of these "export zeros" is strongly correlated with distance and importing country size. Export unit values are positively related to distance and negatively related to market size. The paper concludes with a variant of the heterogeneous-firms trade model pioneered by Marc Melitz (2003). This model maintains the core structure of Melitz, namely heterogeneity in firms' productivity with fixed market entry costs, but introduces a taste for quality so that the lowest priced goods are not necessarily the most competitive. In this model, firms' competitiveness depends upon their quality-adjusted price and, in equilibrium, higher quality goods are more costly, more profitable and better able to penetrate distant markets. The model's predictions are borne out by the facts established in the data analysis. The paper also shows that the export zero prediction is that the probability that no firm exports in a given code from o to d is increasing in the distance between o and d and in the size of d. The export price prediction is that, within a given code, higher-cost products are less likely to be shipped longer distances and to less remote markets. As a consequence of this composition effect, the average export price within a code will be decreasing in distance, increasing in remoteness, and unrelated to size. The paper also shows that the export zero prediction with linear demand is that the probability that no firm exports in code i from o to d is decreasing in the distance between o and d. Unlike in the EK model, the export price prediction with linear demand has nothing to do with a composition effect. Rather, the prediction is driven by the reduction in markups with distance. The paper also shows that the export zero prediction with linear demand is that the probability that no firm exports in code i from o to d is decreasing in the distance between o and d. The export price prediction is that, within a given code, higher-cost products are less likely to be shipped longer distances and to less remote markets