In this work an Unsupervised Neural Computing Model formed by two neural networks is presented: a Self-Organizing Map (SOM) Network and a Hopfield Recurrent Neural Network (HRNN). The first network extracts the endmembers found in the image, analyzing each pixel, and the second network gets the endmember abundances for each pixel in the image. One of the application fields of the proposed methodology is the water quality analysis. In order to study the behaviour of the proposed model, simulation methods have been used to generate hyperspectral signatures from the water spectra obtained in the laboratory. Such data are used for the training and testing of the network. The first sub-network extracts, from the datasets, the endmembers that are used as training patterns in the second one, that provides the matching abundances. The results obtained here will be applied to the treatment of the hyperspectral image Cáceres ES-4, got by the sensors DAIS and ROSIS, from Guadiloba reservoir.