To analyze links among key processes that contribute to Arctic-midlatitude teleconnections we apply causal discovery based on graphical models known as causal graphs. First, we calculate the causal dependencies from observations during 1980–2021. Observations show several robust connections from early to late winter, such as atmospheric blocking within central Asia via the Ural blocking and Siberian High, the North Atlantic Oscillation phase and the polar vortex (PV). The PV is affected by poleward eddy heat flux at 100 hPa, which is also directly connected with the Aleutian Low. We then evaluate climate models participating in the Coupled Model Intercomparison Project Phase 6 (CMIP6) by comparing their causal graphs with those derived from observations. Compared to observations, CMIP6 historical and future simulations do not robustly capture Arctic-midlatitude teleconnections arising from Arctic sea ice variability. This highlights the role of atmospheric internal variability in modulating the Arctic-midlatitude teleconnections. However, we find several distinct patterns that are simulated by most of the analyzed climate models. For example, both historical and future model simulations robustly capture observed atmospheric blocking in central Asia. But contrary to observations, model simulations show a robust link between the Arctic temperature and sea ice cover over Barents and Kara seas. The analysis of future changes also reveals that the connection between the Aleutian Low and the poleward eddy heat flux at 100 hPa is expected to become more robust toward the end of the 21st century than in the analyzed past.