This paper explores the robustness of the Balassa-Samuelson (BS) hypothesis. We analyze an OECD country panel from 1970 to 2008 and compare three data sets on sectoral productivity, including newly constructed data on total factor productivity. Overall, our within- and between-dimension estimation results do not support the BS hypothesis. Over the last two decades, we find a robust negative relationship between productivity in the tradable sector and the real exchange rate, even after including the terms of trade to control for the deviations from the law of one price. Earlier supportive findings depend on the choice of the data set and the model specification.