Two hundred years after von Humboldt's pioneering work on the upper treeline, and many fundamental studies thereafter, the rate of past elevational changes in one of the most fascinating biogeographic boundaries on our planet remains poorly understood. Here, we distinguish conceptually between realised and potential treeline positions and present an ensemble model approach to simulate mean elevational treeline fluctuations for the past 2000 years. Based on dendrochronological summer temperature records, our simple, though efficient model shows that alpine treelines across the Northern Hemisphere were, on average, about 45-50 m higher after the Roman and medieval warm periods compared to their lowest mean position in the 1760s, which likely reflects the accumulated effects of repeated cold spells during the Little Ice Age. We suggest that the simulated mean differences between realised and potential treeline positions can affect the amplitude and variance of tree ring-based temperature reconstructions. Contrary to common belief, we also argue that the current offset between lower realised and much higher potential treeline positions, which appears unprecedented in the past two millennia, does not account for the yet unexplained ‘Divergence Problem’ in dendroclimatology, a decoupling between increasing temperature measurements and tree-ring chronologies since the 1970s.