This study investigates whether a regional climate model (RCM) driven by a global general circulation model (GCM) in a nesting approach with observed atmospheric CO2 concentrations shows predictability for temperature and precipitation trends during 1961–1990 in the Mediterranean area, a region strongly influenced by large-scale circulation. Resulting discrepancies between model and observations raise the question whether the model predictability increases after removing impacts of mid-latitude circulation variability. For temperature and precipitation trends we use the RCM REMO and the observational dataset E-OBS, and for atmospheric circulation the driving coupled GCM ECHAM5/MPI-OM and NCEP/NCAR reanalyses. Cross-validated multiple regression analyses between large-scale circulation and regional temperature and precipitation are performed for observed and simulated data. The impact of circulation is removed from the original temperature and precipitation data, and the trends of circulation-related and circulation-unrelated parts are compared. The circulation-related trends of models and observations show discrepancies owing to differing observed and simulated mid-latitude circulation dynamics, i.e. different temporal evolutions of North Atlantic Oscillation and East Atlantic pattern in winter and East Atlantic Jet and a blocking pattern in summer. Such differences can be related to unknown initial conditions of GCM simulations. In fact, we find strong impacts of initial conditions on mid-latitude circulation dynamics of ECHAM5/MPI-OM ensemble members over 30-year periods. The agreement between simulated and observed circulation-unrelated trends is generally higher than for original trends indicating that the predictability of this nesting approach increases by removing impacts of mid-latitude circulation variability. We conclude that initial conditions affect climate variability up to the multi-decadal timescale, at least in parts of the globe which are governed by extratropical circulation modes, and hence, hinder the comparability of simulated and observed climate trends over time periods shorter than the timescale dominated by radiative forcing. In the Mediterranean Basin the latter is definitely beyond 30 years.