We investigated the effect on the predicted induced field inside the head tissues of the numerical model adopted for simulating the excitation coil during transcranial magnetic stimulation (TMS). A commercially available eight-shaped coil was modeled in four different ways. The electric field inside a detailed (2 mm resolution) and anatomically realistic numerical phantom of the human head was calculated with the impedance method. The results revealed differences of more than 170% in the maximum local value of electric field, depending on the coil model. It was shown that the predicted amount of involved tissue in the brain varied as well, pointing to the importance of correct source modeling before individualized patient treatment planning can be introduced. The calculations showed clearly that an eight-shaped TMS coil should not be modeled as two single thin-wire loops, since this approach may yield inaccurate induced field distributions.