AbstractClimate model simulations are currently the main tool to provide information about possible future climates. Apart from scenario uncertainties and model error, internal variability is a major source of uncertainty, complicating predictions of future changes. Here, a suite of statistical tests is proposed to determine the shortest time window necessary to capture the internal precipitation variability in a stationary climate. The length of this shortest window thus expresses internal variability in terms of years. The method is applied globally to daily precipitation in a 200-yr preindustrial climate simulation with the CMCC-CM coupled general circulation model. The two-sample Cramér–von Mises test is used to assess differences in precipitation distribution, the Walker test accounts for multiple testing at grid cell level, and field significance is determined by calculating the Bejamini–Hochberg false-discovery rate. Results for the investigated simulation show that internal variability of daily precipitation is regionally and seasonally dependent and that regions requiring long time windows do not necessarily coincide with areas with large standard deviation. The estimated time scales are longer over sea than over land, in the tropics than in midlatitudes, and in the transitional seasons than in winter and summer. For many land grid cells, 30 seasons suffice to capture the internal variability of daily precipitation. There exist regions, however, where even 50 years do not suffice to sample the internal variability. The results show that diagnosing daily precipitation change at different times based on fixed global snapshots of one climate simulation might not be a robust detection method.