Incident power density is used as the dosimetric quantity to specify the restrictions on human exposure to electromagnetic fields at frequencies above 3 or 10 GHz in order to prevent excessive temperature elevation at the body surface. However, international standards and guidelines have different definitions for the size of the area over which the power density should be averaged. This study reports computational evaluation of the relationship between the size of the area over which incident power density is averaged and the local peak temperature elevation in a multi-layer model simulating a human body. Three wave sources are considered in the frequency range from 3 to 300 GHz: an ideal beam, a half-wave dipole antenna, and an antenna array. 1D analysis shows that averaging area of 20 mm × 20 mm is a good measure to correlate with the local peak temperature elevation when the field distribution is nearly uniform in that area. The averaging area is different from recommendations in the current international standards/guidelines, and not dependent on the frequency. For a non-uniform field distribution, such as a beam with small diameter, the incident power density should be compensated by multiplying a factor that can be derived from the ratio of the effective beam area to the averaging area. The findings in the present study suggest that the relationship obtained using the 1D approximation is applicable for deriving the relationship between the incident power density and the local temperature elevation.