The most recent safety guidelines define basic restrictions for electromagnetic field exposure at frequencies more than 6 GHz in terms of spatial‐ and time‐averaged transmitted power density inside the body. To enable easy‐to‐perform evaluations in situ, the reference levels for the incident power density were derived. In this study, we examined whether compliance with the reference levels always ensures compliance with basic restrictions. This was evaluated at several distances from different antennas (dipole, loop, slot, patch, and helix). Three power density definitions based on integration of the perpendicular real part of the Poynting vector, the real part of its three vector components, and its modulus were compared for averaging areas of λ2/16, 4 cm2 (below 30 GHz) and 1 cm2 (30 GHz). In the reactive near‐field (d < λ/(2π)), the transmitted power density can be underestimated if an antenna operates at the free space exposure limit. This underestimation may exceed 6 dB (4.0 times) and depends on the field source due to different coupling mechanisms. It is frequency‐dependent for fixed‐size averaging areas (4 and 1 cm2). At larger distances, transmission can be larger than the theoretical plane‐wave transmission coefficient due to backscattering between the body and field source. Using the modulus of the incident Poynting vector yields the smallest underestimation. © 2020 Bioelectromagnetics Society.