Radio frequency sources heavily impact human life due to the high number of technologies exploiting them. Their level assessment is a complex task because of differences in technologies working principles and high growth rate. Among them, cellular communications are the most pervasive, and several concerns are often addressed to currently deploying 5G. Here, 4G long-term evolution (LTE) technology is focused because it is still the largest used communication facility. To assess human exposure due to LTE, apart from traditional broadband and narrowband methods, easier and faster approaches have been proposed in the literature and the most recent technical standards: the extrapolation techniques (ETs). They measure pilot signals' levels and process them to obtain the equivalent worst case channel power. They are designed to overestimate the current channel power by estimating the maximum channel power in the measurement point, thus warranting a conservative approach. Nevertheless, some of the needed hypotheses, such as the pilot signal power constancy, do not always pass the experimental validation. Based on that, in some cases, supposed overestimation can turn into underestimation, thus losing the "conservative" feature and making the power measurement unreliable. Through a wide experimental analysis, this article aims to highlight those issues and derive an improved measurement procedure, under the assumption that a basic spectrum analyzer is adopted as a measuring instrument.