The correct depiction of atmospheric blocking still poses a key challenge for current numerical weather prediction (NWP) and climate models. This study evaluates the representation of blocking in the new global ICOsahedral Non-hydrostatic NWP and climate model ICON and links model mean state biases to observed blocking deviations. Blocking is identified using both an anomaly and a flow reversal approach in an eight member ensemble of 15-year AMIP-type ICON simulations and verified against ERA Interim reanalyses. Either approach demonstrates a good representation of annual blocking frequencies in ICON. Deviations emerge when considering individual seasons. In the anomaly framework, enhanced blocking occurrence in the mid-latitude Pacific domain during winter and spring and a marked underestimation of blocking in the Euro-Atlantic region are found during summer. Moreover, this approach indicates a general underestimation of blocking at higher latitudes. The flow reversal index reveals the often reported underestimation of blocking in the Euro-Atlantic region during winter. Furthermore, increased blocking activity in the Pacific and Greenland region during spring and decreased blocking occurrence at high latitudes in summer are found. Focusing on the anomaly approach, we assess how the model mean state influences blocking identification. A systematically higher tropopause, forced by a cold bias in the lower stratosphere, reduces diagnosed blocking frequencies at higher latitudes especially during summer. This goes along with a reduction in blocking size, duration, and intensity. While confirming an overall good representation of blocking in ICON, this study demonstrates how mean state biases can crucially affect the identification of blocking and that blocking deviations have to be interpreted with caution as they are highly dependent on the exact diagnostic used.