Glow Discharge Optical Emission Spectroscopy (GD-OES) is a well established technique for Compositional Depth Profiling (CDP), very useful for in-depth elemental analysis of surface layers from 10 nm to 100 µm. It is fast, sensitive and fully quantitative. However, there are a few artefacts requiring further study to improve the accuracy and trueness of the method further, primarily for near-surface applications. Outgassing of volatile compounds present inside the source, mainly water and hydrocarbons, has been established to be a major cause of such artefacts. The most “problematic” element is hydrogen, for the following two reasons: 1) it “cools” the plasma even when present in small quantity and thereby affects the intensities of the emission from almost all other elements; 2) it is very reactive and forms molecular species with other light elements e.g. CH, NH and OH, having emission spectra overlapping several atomic analytical lines; causing “false” or exaggerated surface peaks of the corresponding elements. In this work, methods to reduce the outgassing of volatile compounds have been studied. Attempts to dry and clean the argon gas flowing into the source by means of a chemical filter gave no significant improvement. On the other hand, replacing the “standard” anode made of a copper-beryllium alloy with a pure copper anode was shown to reduce the outgassing significantly. To be more specific, the background signal from atomic hydrogen and associated artefacts are reduced, it is still possible that the amount of molecular hydrogen remains almost constant, without observable effects on the analytical signals. Since molecular emission can originate from the sputtered material of the sample itself, particularly organic coatings, an in-depth study of molecular emission in GD-OES was carried out. It was shown that background signals from such emission can be significantly reduced, provided that the instrument has spectral channels for the emitting molecular species installed. With such channels available, conventional “line interference correction” methods was found to be effective in reducing false elemental signals from molecular emission. Application of the pure copper anode and reduction of molecular background techniques to a heat treated zinc-base coating showed expected improvements in the near-surface part of the depth profile. However, for the major elements of technical importance, the difference compared with the original anode and analytical method was marginal. This is reassuring, since it means that the near-surface artefacts normally do not cause major analytical errors. The element that is most difficult to establish the “trueness” of depth profiles is nitrogen, especially if the surface layer is slightly porous. There are samples where an elevated signal from nitrogen in the top surface cannot be correlated to a vacuum leak, trapped air or molecular emission (CO). Investigations where samples are measured for total nitrogen with conventional techniques are very difficult, since the surface layers showing an elevated content are very thin. Further investigation of such samples with high vacuum techniques, e.g. SIMS, would be of interest. A second part of the project deals with advanced evaluation of depth profile data by means of “expert systems”. An expert system is a computer application that is able to perform tasks which are normally performed by human experts; in this case the aim is to be able to perform e.g. quality control with GD-OES without the need for a qualified human expert to interpret the data. There are several types of “computational statistics” methods that can be employed for such purposes, the most well known is probably multivariate analysis. Other methods can evaluate both numerical and other types of input related to classification based on technical properties, e.g. paint adhesion, scratch resistance etc. The objective of this work is to be able to classify samples according to such technical properties. All such systems need a “training set” of samples with known technical properties. In this work, a relatively large set of steel sheet samples with various zinc-based coatings were classified according to corrosion resistance. Form the GD-OES depth profiles, the coating weights of zinc, aluminium and magnesium were extracted in a data pre-treatment step. Using the rule-based classification algorithm “decision tree” 25 out 29 samples was correctly classified. This “proof of concept” work has shown that it is possible to predict certain technical properties based on a multi-element depth profile. This opens up the possibility for e.g. automated quality inspection of complex coating systems, but also the possibility to use GD-OES depth profiling more effectively as a tool in product development.