A mechanism for automatic detection, identification and compositional quantification of elements in an EELS spectrum image is described. The method is capable of locating elemental occurrences, discovering signal overlaps, correctly modeling and subtracting the background, or alternatively fitting reference spectra to each image pixel to convert image intensities at any point in a spectrum image to a concentration without almost any operator input, thus paving the way for a completely automated spectrum image analysis. We describe the steps involved in extracting the elemental content in a spectrum image and demonstrate how an image can be derived that clearly reveals the problem zones that prevent accurate results in a subsequent quantification. Such an automatically generated image can then serve as a binary mask, which allows performing selective calculations on certain specimen areas, when applied to the original data set. We demonstrate the feasibility of such an approach by displaying examples computed from ceramic as well as alloy and steel specimens.