Neural networks are simplified mathematical models to simulate certain aspects of information processing in biological nervous systems. Some principles adopted from nature include parallel processing, learning by example, and abstraction of knowledge. Neuro-computing is a rapidly growing branch of research, where constantly new types of neural networks are developed for different applications. Pattern recognition and classification is the typical application of such approaches. Many tasks in geosciences, particularly in data interpretation, can be considered and treated as classification problems. In such a terminology, the studied objects (e.g. sub-regions of earth models) are classified as rock types based on characteristic features (e.g. physical properties). Numerical solutions have to address challenges such as incorrectness and incompleteness of data, and overlap in some properties for different lithologies. The self-organizing map is an intriguing concept, which allows to establish a certain classification behavior by unsupervised learning. We developed a work flow at GFZ, which includes preparation of data, application of learning rules, the segmentation of the trained map by adopting image processing techniques, and employment of knowledge. Applications are shown from geothermal exploration projects in the Northeast German basin and in Indonesia, where different geophysical models from the same study area were combined for a joint lithological interpretation.