Multi-frequency microwave radiometry has been considered for the measurement of sub-surface temperature distributions of human tissues, envisaged applications being mainly related to the non-invasive temperature monitoring in hyperthermia treatment of cancer. The sensor consists of a broadband microwave receiver, which generates a signal proportional to input radiation power, then a calibration processor averages the receiver output and computes an estimate of the input power in each frequency band. We have developed a new radiometer system of the radiation balance type to counteract the reflectivity errors which may affect the multi-frequency radiometric data to an intolerable amount. In this paper we discuss the reflectivity error of our system on the basis of a model of both incoming radiation and receiver. A formula, which propagates this error to temperature retrieval uncertainty, is also given.