We calculated the whole-body-averaged specific absorption rates (WBSARs) in a Japanese 9-month-old infant model and its corresponding homogeneous spheroidal and ellipsoidal models with 2/3 muscle tissue for 1–6 GHz far-field exposure. As a result, we found that in comparison with the WBSAR in the infant model, the ellipsoidal model with the same frontally projected area as that of the infant model provides an underestimate, whereas the ellipsoidal model with the same surface area yields an overestimate. In addition, the WBSARs in the homogenous infant models were found to be strongly affected by the electrical constant of tissue, and to be larger in the order of 2/3 muscle, skin and muscle tissues, regardless of the model shapes or polarization of incident waves. These findings suggest that the ellipsoidal model having the same surface area as that of the infant model and electrical constants of muscle tissue provides a conservative WBSAR over wide frequency bands. To confirm this idea, based on the Kaup index for Japanese 9-month-old infants, which is often used to represent the obesity of infants, we developed linearly reduced 9-month-old infant models and the corresponding muscle ellipsoidals and re-calculated their whole-body-averaged SARs with respect to body shapes. Our results reveal that the ellipsoidal model with the same surface area as that of a 9-month-old infant model gives a conservative WBSAR for different infant models, whose variability due to the model shape reaches 15%.