The Atlantic Meridional Overturning Circulation (AMOC), a crucial element of the Earth's climate system, is projected to weaken over the course of the twenty-first century which could have far reaching consequences for the occurrence of extreme weather events, regional sea level rise, monsoon regions and the marine ecosystem. The latest IPCC report puts the likelihood of such a weakening as ‘very likely’. As our confidence in future climate projections depends largely on the ability to model the past climate, we take an in-depth look at the difference in the twentieth century evolution of the AMOC based on observational data (including direct observations and various proxy data) and model data from climate model ensembles. We show that both the magnitude of the trend in the AMOC over different time periods and often even the sign of the trend differs between observations and climate model ensemble mean, with the magnitude of the trend difference becoming even greater when looking at the CMIP6 ensemble compared to CMIP5. We discuss possible reasons for this observation-model discrepancy and question what it means to have higher confidence in future projections than historical reproductions. This article is part of a discussion meeting issue 'Atlantic overturning: new observations and challenges'.