The paper considers the problem of estimating the parameters of the compound admittance matrix of three-phase untransposed low-and medium-voltage electrical distribution networks using synchrophasor measurements from phasor measurement units (PMUs). The work proposes and analyses the performance of a pre-processing strategy on the PMU’s raw measurements which consists in grouping the raw measurements in clusters and, then, using the averaged measurements from each cluster for admittance matrix estimation. This step reduces the noise level and discards similar measurements from each cluster, ultimately improving the estimation quality of regression-based estimation methods such as least squares (LS) and total least squares (TLS). The proposed approach uses a linear estimation model with phasor measurements of nodal voltages, nodal injection currents, and branch currents. We adopt a realistic measurement noise model in polar coordinates, which reflects the accuracy class of measuring instruments and is projected to rectangular coordinates. The proposed approach is validated by performing simulated experiments for different CIGRE and IEEE benchmark grids. Furthermore, the work includes different sensitivity analyses on the pre-processing policy (cluster type and size), availability of branch or injection currents measurements, as well as with different noise levels on the measurement data.