The paper considers the task of evaluating product quality indicators based on the results of measurements obtained during the control testing. Data processing is proposed to be carried out in two stages. First stage comprises data compression via cluster analysis, and the second stage uses the procedure of non-parametric estimation of the observed measurements to evaluate the quality of small samples with an unknown distribution law. Quality scores are defined as guaranteed scores on a set of distributions with moments equal to sample points found from a small sample. A number of theoretical statements are formulated, and a model example is given.