If the main purpose of a sampling survey consists in the estimation of a change (from time t-k to time t) concerning a certain quantitative variable y, the presence of outliers could lead to significant biases in the estimation process. In this paper we’ll deal with the first problem, since we adopted the rule that if a unit is identified as outlier, we exclude it from any further calculation, so we assign to it a weight equal to zero. Starting from a lightly improved version of a basic data transformation originally proposed by Hidiroglou and Berthelot, we present a new technique for detecting outliers (minimum variance procedure), that generally leads to a significantly lower amount of units identified as outliers and of micro-data reject. Moreover, we propose various qualitative indicators to assess the optimality of any outlier detection method. Finally, starting from the availability of monthly data concerned with the retail trade monthly survey currently carried out by ISTAT, we compare nine methods for identifying outliers, drawing some main conclusions on the optimality of procedures based on the proposed minimum variance procedure.