Safety models on scanners are unaware of the actual patient position while excitation pulses are inherently position dependent. This study investigates the effect of this positional mismatch on SAR estimation for axial, coronal and sagittal slice orientations. The positional mismatch yields up to 5.2-fold underestimation of peak local SAR. RF shimming and 2-spoke parallel-transmit pulses for axial slice orientation have reduced SAR-sensitivity to positional mismatch with the worst-case underestimation being <2.0-fold whereas a reduced SAR-sensitivity was not observed for coronal and sagittal slices. For extreme head positions not represented in safety-models, axial RF shimming / 2-spokes parallel-transmit pulses maybe beneficial.