The optimisation of parameters is investigated for the compositional analysis of nanometre-sized particles embedded in a matrix by energy-filtered transmission electron microscopy. The specific example of Cu-rich particles in a Fe matrix is used both to model and to explore the experimental limits of detection and characterisation. Modelling of alternative procedures for background extrapolation as a function of the number of pre-edge windows confirmed that greater accuracy in a fixed analysis time is achieved by using more than two pre-edge windows. Further modelling investigated the effects of noise, drift and instrumental blurring of images on the accuracy of particle size and composition measurements. Correction factors were generated for ranges of these artefact amplitudes. The corrections were then applied to experimental data and shown to be both realistic and effective. Determination of particle radius below 1 nm was demonstrated.