'The focus of this thesis is to develop a credit scoring model that, with a higher degree of certainty than earlier, will obtain a greater share of those customers that meets the conditions defining solvency. The model is constructed through an iterative process composed of two procedures: First a selection-model is created based on information from solvent customers only. In this manner I make use of information on customers, who in the past had shown to be creditable, and argue that those have qualities that the bank in the feature should opt for. Secondly the model is modified to make sure it accepts a certain share of applicants. This must be done to avoid the reject inference problem (because the model is based on solvent customers only) and to avoid that the selection mechanism over time would accept still fewer applicants. To ensure I get the topmost share of applicants I choose a classification method that can tell how certain it is on its conclusion. This led to selecting a modified Naïve Bayes classifier, altered in such a way that it accepts a specific share of applicants. The method was tested using a virtual experiment and the result indicated that the method has some good promise. There are however issues that should be investigated more closely, e.g. the construction of the experiment and alternatives to the Naïve Bayes classifier.'