This work describes a new method of composition and thickness determination in high resolution electron microscopic samples by quantitative image processing. The method performs a lateral resolution of 5.9 Angstroems. In this work only III-V semiconductors (sphalerite structure) are investigated though the used methods can be generalized easily to other material systems. One of the dominant source of error in the quantitative high resolution electronmicroscopy is surface amorphisation during sample preparation. Such surface amorphization was simulated in a semianalytical model. By comparison with HREM images the influence of surface amorphization is investigated. The use of neural networks for this application was motivated by function approximation theory als well as through comparison with classical methods. By application of this method to HREM images of AlGaAs heterostructures random alloy fluctuations in the ternary semiconductor were quantitatively shown and compared to a theoretical model. Determination of crystalline sample thickness shows thickness variations due to sample preparaton.