The article analyzes clustering problems that arise in forecasting tasks when clustering short time series of demand data. During the clustering process each object is assigned to a particular class or group of objects. The use of the same clustering algorithm does not guarantee results that would have the same interpretation if there are changes in clustering error evaluation method or the number of clusters is different. Robust clustering shows the differences in the object allocation to groups or clusters that occur when the clustering algorithm is repeatedly run using the same data set. The article provides a short description of the used methods and algorithms that are implemented in the process of data clustering and robust clustering analysis. The goal of this article is to perform the analysis of results with ambiguous interpretations and to give recommendations for solving similar problems. The clustering is carried out using k-means algorithm and its modification as well as clustering algorithms used in the data mining tools Weka, Orange Canvas and the statistics package Statistica. The distance calculations implement Euclidean distance. The robust clustering analysis is performed using Calinski-Harabasz approach that selects the number of clusters as an argument value in a function that is being maximized. Comparative analysis of the obtained results is carried out using a unified data set of product demand values that are defined by short time series with a fixed period of time. The conclusions of the article reflect authors’ recommendations that had arisen as a result of the research.