Several recent papers have underlined the importance of the microstructure effects in understanding exchange rate behavior by documenting stable long-run relationships between cumulated order flows and spot exchange rates. This stands in contrast to the widely-studied failure of exchange rates to conform to the long-run behavior implied by “conventional” macroeconomic models and is consistent with the prediction of micro-structure models. We reexamine the evidence for stable long-run relationships. We find that such evidence exists only for a small number of the major currencies we examine and that is it statistically fragile. We conclude that this implication of microstructure models does not fit the data as well as previous studies suggest.