The properties of stainless steels can be significantly influenced by the cleanness level. Through Electroslag Remelting (ESR) a considerable improvement of steel cleanness is achieved. This refining effect results from the contact between the liquid metal and the reactive slag bath at several reaction sites. Regarding cleanness improvement, the understanding of reactions at the steel/slag interfaces is indispensable. Within the present work, the modification of oxides in the steel X38CrMoV5-1 through remelting is studied from the viewpoints of experiments, thermodynamics and kinetics, especially focussing on the spinel inclusion type MgOAl2O3. Next to a detailed investigation of samples from the ESR process, laboratory experiments in a Tammann Furnace are performed. Subsequent SEM/EDS analyses provide a detailed insight into the size, the composition as well as the morphology of inclusions. Thermodynamic considerations are addressed by the software FactSage. Additionally, a Laser Scanning Confocal Microscope enables the in-situ observation of the dissolution behaviour of oxides in fluoride-slags. Since the reliable characterisation of non-metallic inclusions is of major importance for this study, the representativeness of automated SEM/EDS analysis is affirmed for the defined conditions by coupling a geometric-statistical model with experimental considerations. Furthermore, the limitations of this method are critically reviewed. This thesis especially concentrates on the changes in inclusion morphology which give a strong indication regarding the modification mechanisms during remelting. Next to the determination of the most decisive reaction site, the influence of the electrode quality on the final inclusion landscape in the remelted ingot is assessed. Moreover, the significance of slag composition and dwell time of inclusions in the liquid pool is discussed. The findings provide important knowledge about the essential influencing factors on the overall removal rate of inclusions through Electroslag Remelting and therefore for further process optimisation.