I present an update of the single-station polarization filter proposed by Jin and Ronen (2005) for ground-roll elimination. This process is carried out by treating a sliding time window of a single-station triaxial recording as a matrix and doing singular value decomposition of this matrix. The efficiency of this filter depends on how reliably the ground roll is detected. For this reason, I derive an attribute to be minimally sensitive to the corruptive impact of additive random noise and, on the other hand, maximally sensitive to the presence of the ground roll. The advantage of using this attribute is demonstrated on 3C field data.