The atmospheric Integrated Forecast System model from the European Centre for Medium-Range Weather Forecasts is used to calculate forcing perturbations which are optimal in producing atmospheric response patterns resembling the ‘cold-ocean-warm-land’ (COWL) flow regime over 4 d. Similar initial state perturbations are computed for comparison. COWL is relevant for recent climate change. The perturbations are optimal in a tangent-linear sense but are validated by non-linear calculations. Calculations cover 836 cases during 22 winter seasons. The method effectively estimates flow-dependant perturbations which produce response patterns resembling COWL. The optimal forcing is more geographically confined with relatively smaller remote amplitudes and larger spatial scale than the initial state perturbations. The quality of the 4-d response is highly dependent on the actual optimal perturbation. Averaging over just a few cases drastically reduces this optimal property, but the long-term climate response of an average forcing does produce signatures of the COWL regime. The results are discussed in view of Palmer's non-linear dynamical perspective on climate change, and several key elements are confirmed: the climate sensitivity is flow dependent; efficient forcing structures do not resemble the response (non-normality); sensitivity and predictability are negatively correlated; and flow characteristics for high sensitivity differ from those for low sensitivity.