In the agro-food industry, determining the quality of products is of utmost importance. Among the numerous analytical techniques available, rapid and non-destructive instrumental methods, such as Near Infrared spectroscopy, are often preferred. However in order to extract information from the acquired spectral data, chemometrical methods (i.e., mathematical or statistical methods) are nearly always required. PLS_Cluster was proposed by Barros et al. [A.S. Barros and D.N. Rutledge 70:99 2004] as an unsupervised clustering method, which, being based on PLS regression, furnishes interpretable B-coefficients for each node in the dendrogram. In the present paper a generalisation of the original PLS_Cluster method is proposed. The original version of PLS_Cluster was dichotomic, the samples being assigned to just two groups at each node. As well, the membership function used for the assignment of the samples was not always optimal. This new algorithm yields more satisfactory results, and has the additional advantage over other clustering methods of being predictive. A further extension of Generalised PLS_Cluster is also proposed, i.e. Interval PLS_Cluster, in which the PLS_Cluster algorithm is applied on a window moving across the spectra. This may result in different clusterings of the samples for different spectral regions and enables the determination of those variables responsible for the observed clusterings. Application of Generalised- and Interval- PLS_Cluster to several data sets gave satisfactory results.