Debt collection is a massive industry, within the USA alone more than $50 billion recovered each year. However the information available is often limited and incomplete, and predicting whether a given debtor would repay is inherently a challenging task. This has amplified research on debt recovery classification and prediction models of late. This report considers three main mathematical, data mining and statistical models in debt recovery classification, in logistic regression, artificial neural networks and affinity analysis. It also compares the effectiveness of the above-mentioned tools in evaluating whether a debt is likely to be repaid. The construction and analysis of the models were based on a fairly large unbalanced data sample provided by a debt collection agency. We have shown that all three models could classify the debt repayments with a considerable accuracy, if the assumptions of the models are satisfied.