The invention relates to a method of reducing the variance of a signal-to-noise ratio estimate for a signal with differential phase and coherent amplitude modulation. The invention is characterised in that, in order to calculate the samples of the noise power affecting the system, the amplitude and phase increment errors obtained at the receiver are combined, by shifting said errors (1) and (4) to the corresponding point of the first ring (6) of the constellation as if said ring had been transmitted and without undergoing any modification, in order subsequently to calculate the square of the modulus of the error vector (9). In this way, the invention minimises and equalises the variance of the samples obtained for the different even and odd values of bits for each carrier used by the system. In addition, the invention reduces the variance of the signal-to-noise ratio estimate in a multi-user digital communication system. The aforementioned estimate is necessary for selecting the number of bits per carrier, such that same can be adapted to the channel, as well as being necessary as support information when a diversity receiver is used. (FIG. 4)