Europe's forests play a critical role as carbon sink, timber production and for the societal well-being, yet their capacity to maintain these functions is increasingly threatened by increasing disturbance rates and a growing demand for wood. Reliable data on disturbance-induced tree cover change are essential to address the challenges faces by Europe's forests. Existing Earth Observation-based products, however, have largely unquantified uncertainties, limiting their potential for robust estimation and reporting. Robust estimation of tree cover change rates and trends, as well as of map uncertainties, requires reference data on tree cover change, covering all of Europe using a probabilistic sample and a harmonized response design. Here, we address this gap by developing a new database of forest disturbances for Europe, based on consistent manual interpretation of satellite imagery from the Landsat archive. Our database covers 35,000 sites across 35 countries, with each site being a 900 m<sup>2</sup> square corresponding to a Landsat pixel, yielding a total of 1,365,000 individual site-year observations. Interpretation was done over the period 1985 to 2023 by trained interpreters, who – for each site-year combination – determined whether a disturbance occurred or not. Disturbances were thereby defined as tree cover loss detectable at a spatial resolution of 900 m<sup>2</sup>, caused by either human activity (i.e. harvest), natural disturbances (e.g. bark beetle, fire, wind, avalanches, flooding, etc.) or both (e.g. bark beetle with following salvage logging). Using our database, we exemplify the value of manually interpreted reference data by deriving robust sample-based estimates of disturbance rates and trends. We further compare our sample-based estimates to state-of-the art map-based estimates, highlighting uncertainties in relying solely on Earth observation. Our results underscore the need for consistent, transparent, and independent reference data for understanding disturbance change in Europe and for validation and uncertainty quantification in Earth observation data.