Standard tests of the Harrod-Balassa-Samuelson (HBS) hypothesis treat productivity levels in and across countries as fixed and observable, and offer little empirical support for the hypothesis. If productivity follows a jump-diffusion process, these standard tests will generate biased estimates, measuring productivity levels with error. This paper instead proposes an ‘errors in variables’ approach to correct this bias, and finds support for the HBS hypothesis assuming a jump-diffusion process in productivity. Empirical results are obtained for a data set available for the United States, Japan, West Germany and France over the period 1960 to 1996.