We investigate whether real exchange rates under the gold standard can be modeled as stationary around a broken trend. Using both conventional unit root and sequential trend break tests, we clearly reject the unit root null for 14 of the 16 exchange rates studied by Diebold, Husted and Rush (Journal of Political Economy, 1991, 99) and provide some evidence against a unit root for another. We interpret our results as being complementary to the evidence for fractional integration found by Diebold, Husted and Rush. In both cases, the unit root null can be decisively rejected if the alternative is more flexible than a deterministic trend without breaks. © 1995.