The sensitivity of RF CMOS receivers using a direct conversion or a low-IF architecture is strongly affected by flicker noise. This paper gives theoretical guidelines to predict the flicker noise in Gilbert-cell mixers. The conversion gain, the equivalent input and output noise, and the effect of the pole at the single internal RF node are discussed. For the first time, results which are valid in all modes of operation are given. Such complete results are required for some ultra low-power and low-voltage applications, since the transistors might be operated in moderate or even weak inversion region. The theoretical gains are found to remain within a 2-dB margin with respect to the measurements of a UHF downconverter built in a 0.5-μm process, for a large range of bias conditions and local oscillator swing.