The paper presents a novel approach to reveal loads having the biggest impact on damping of small-disturbed oscillations in power systems. The technique is based on a quasi-optimisation procedure with the cost function reflecting shifts of selected eigenvalues along the real axis when all unknown load parameters are varied within their constraints. The cost function takes into consideration a variety of power system operating conditions. All points of the cost function steepest descent trajectory, obtained in the space of unknown load parameters, correspond locally to biggest shifts of eigenvalues. Participation factors of load parameters are computed along the steepest descent trajectory, and they give ranks of load importance. Small ranks indicate loads which have no influence on damping of selected modes. Big ranks reveal loads whose parameters cause the biggest change in damping under the most unfavourable combination of unknown parameters. Parameters with biggest ranks should be measured first of all to avoid mistakes in evaluation of damping properties and power system stability margins.