The exposure of children to mobile phones has been a concern for years, but so far the conclusions with respect to compliance with safety standards are based only on simulations. Regulators have requested that these conclusions be supported by experimental evidence. The objectives of this study are 1) to test if the hypothesis that the specific anthropomorphic mannequin (SAM) used in standardized compliance testing is also conservative for homogeneous child head models and 2) to validate the numerical prediction of the peak spatial SAR (psSAR) in child head phantoms. To achieve these objectives, head phantoms of 3- and 8-year-old children were developed and manufactured. The results confirm that SAM is also conservative for child head phantoms, and that the agreement between numerical and experimental values are within the combined uncertainty of 0.9dB, provided that the actual peak spatial SAR (psSAR) is determined. The results also demonstrate that the currently suggested numerical SAR averaging procedures may underestimate the actual psSAR by more than 1.3dB and that the currently defined limits in terms of the average of a cubic mass are impractical for non-ambiguous evaluations, i.e., for achieving inter-laboratory repeatability.