Unsupervised clustering techniques are a valuable source of information for determining how to divide a dataset into subgroups. We present a comprehensive analysis of the quality of these algorithms by defining a clustering fuzziness metric. A statistical test and cluster probabilities corrections are provided based on this metric. Some examples demonstrate how it can be used to compare different clustering algorithms or improve the accuracy of various methods. An application for adjusting the number of clusters is also presented. These results are illustrated using both simulated and real-world data.