The aim of this paper is to provide a detailed investigation of trade barriers by examining a cross-country, cross-industry analysis of evolution of trade costs for a 44-year period (1963-2006) positing the possible reasons behind the changes. I estimate trade costs using methodology of Shikher (2010), which is a framework based on the EK model (Eaton and Kortum, 2002) at the industry level. Instead of using the Armington assumption (1969) to explain intra-industry trade, the model imposes the producer heterogeneity assumption. Therefore, the goods are di¤erentiated according to their features, rather than their country of origin. Moreover, in case of producer heterogeneity, countries do not have monopoly power, and home bias in consumption and price di¤erence across countries are explained by trade costs rather than demand-side parameters. Another advantage of the model is that it captures trade in intermediate goods, which is important since trade in intermediate goods affects trade costs. The results show that trade costs are large and vary significantly across goods, but the trends are downward for the total manufacturing industry and for the majority of the individual industries. Trade costs of non-OECD countries for all sectors are larger than those of OECD countries, and there is divergence between the trade costs until the 1980s. After the 1980s, there is steady catching up in trade costs of capital goods industries, labor intensive industries and the total manufacturing industry. But differences in average trade barriers are still more than 40% for these sectors. The results according to country group of the trading partner show that, for the total manufacturing industry, if exports come from the OECDs, the barriers to trade are lower than in the case of non-OECD exporters. At the sector level, on the other hand, the trade costs among non-OECDs are lower than the barriers to trade from OECDs to non-OECDs. I also examine which determinant has the largest effect on trade costs, and encounter endogeneity bias issue, which I correct using the differenced panel data approach.