The global analysis of metabolites can be used to define the phenotypes of cells, tissues or organisms. Classifying groups of samples based on their metabolic profile is one of the main topics of metabolomics research. Crisp clustering methods assign each feature to one cluster, thereby omitting information about the multiplicity of sample subtypes. Here, we present the application of fuzzy K-means clustering method for the classification of samples based on metabolomics 1D (1)H NMR fingerprints. The sample classification was performed on NMR spectra of cancer cell line extracts and of urine samples of type 2 diabetes patients and animal models. The cell line dataset included NMR spectra of lipophilic cell extracts for two normal and three cancer cell lines with cancer cell lines including two invasive and one non-invasive cancers. The second dataset included previously published NMR spectra of urine samples of human type 2 diabetics and healthy controls, mouse wild type and diabetes model and rat obese and lean phenotypes. The fuzzy K-means clustering method allowed more accurate sample classification in both datasets relative to the other tested methods including principal component analysis (PCA), hierarchical clustering (HCL) and K-means clustering. In the cell line samples, fuzzy clustering provided a clear separation of individual cell lines, groups of cancer and normal cell lines as well as non-invasive and invasive tumour cell lines. In the diabetes dataset, clear separation of healthy controls and diabetics in all three models was possible only by using the fuzzy clustering method.