Annular dark-field (ADF) imaging by scanning transmission electron microscopy (STEM) is a common technique for material characterization with high spatial resolution. It has been reported that ADF signal is proportional to the nth power of the atomic number Z, i.e., the Z contrast in textbooks and papers. Here we first demonstrate the deviation from the power-law model by quantitative experiments of a few 2D materials (graphene, MoS2 and WS2 monolayers). Then we elucidate ADF signal of single atoms using simulations to clarify the cause of the deviation. Two major causes of the deviation from the power-law model will be pointed out. The present study provides a practical guideline for the usage of the conventional power-law model for ADF imaging.