The aim of this project was to investigate the potential of applying computational statistics (“data mining”) methods to PDA – OES data to find correlation to defects in the final products as well as changes of parameters during the steelmaking process. The computational methods used were multivariate (MVA) data analysis and the rule-based methods “decision trees” and “neural nets”. The project was carried out in close cooperation with Outokumpu Stainless Avesta and the Royal Institute of Technology (KTH). PDA – OES data from sheet samples with and without surface defects were processed by all three methods, all showing statistically significant and consistent correlation. The formation of surface defects is positively correlated to the number of “medium” to “large” inclusions of the classes AlCaMg and AlCa i.e. mixed oxides of these elements. The second part of the research was an evaluation of variations in inclusion characteristics in different process stages using synthetic slags. A large number of samples were collected from experimental trials with two types of synthetic slags, and the conventional process without slag addition for reference. The samples were taken in the ladle furnace at three stages, and in the tundish in connection with the final test sample before casting. PDA – OES data from these samples were evaluated with MVA and decision tree methods. The results showed that the different process stages can be identified from the PDA-OES data with rather good certainty. No significant difference between the use of synthetic slags and the conventional process was detected. In the course of the evaluation work, it was also found that the computational statistics methods must be used with caution. The reason is that data due to “statistical noise” can be identified as significant, giving misleading results. Further work to reduce this problem will be necessary.