Elemental analysis of inhomogeneous materials can now routinely be per
formed at the nanometer level, using characteristic EELS signals. Majo
r progress in spatial resolution and accuracy in quantification has re
cently been fostered by the practical implementation of the spectrum-i
mage mode in a FEG-STEM environment. The required hardware and softwar
e have been elaborated to record, store and process these large amount
s of data. In the present contribution we describe the routines which
have been implemented for extracting quantitative elemental maps from
spectrum-images: (i) the standard background subtraction method for wh
ich the availability of several hundreds of energy loss channels acros
s the edge to be quantified reduces the errors and bias in background
modelling and extrapolation; (ii) a non-negative multiple-least-square
s routine for fitting the experimental spectrum acquired for each pixe
l with a linear combination of reference edges, if possible recorded d
uring the same scan. The impact of these new tools is demonstrated in
a series of situations encountered in materials science (composite mat
erials, metallic multilayers) which all require nanometer resolution a
nd accurate data processing of complex spectra with overlapping edges.