The thesis addressed the effect of a varying MgO grade in MgO-C refractories on both their behavior in contact with a steel melt and the resulting effect on the non-metallic inclusion (NMI) population in the solidified steel. For this purpose, immersion tests were conducted in a semi-industrial steel casting simulator. In addition, the effect of the steel melting process parameters on the NMI population was thoroughly investigated, providing a guideline for the result interpretation for future experiments in the steel casting simulator. Here, a fundamental concept of data evaluation for the NMI characterization in a steel matrix using automated feature analysis was developed. The main NMI types detected in the solidified steel samples were alumina and MnSi-based inclusions. Their number density depended on the steel melt's temperature and amount of dissolved oxygen. A lower MgO grade refractory specimen in contact with the steel melt resulted in a higher proportion of low melting phases on its surface compared to a higher MgO grade specimen. These low-melting phases promoted the formation of MnSi-based inclusions and triggered NMI agglomeration leading to the formation of large alumina inclusions.