Abstract The results of grain size distribution of loess units are suitable for the study of past climate change. There are many methods for the analysis of the grain size, but clustering methods have not been used before. In this article, a new approach in the sediment research, the k-median method was used. This method separates the grain size classes of the youngest loess units of three loess-soil profiles in Poland: Biały Kościół, Złota and Tyszowce. In the analysis, only the content for the loess silt fraction (range from 4 to 63 µm) was used. This allowed eliminating the admixtures of sandy and clay fraction, considered as secondary material. For the purposes of this analysis, silt fractions were divided into smaller units using a 1/4 phi interval, resulting in 16 new subfractions. The grain size data from three profiles were combined into one data set. As a result, four classes with a similar internal of grain structure were obtained. The separated classes showed the sedimentological variability of the sediment well. It has been shown that the k-median method can be used to classify loess in a fast and objective way.