Credit scoring techniques are used to estimate the probability of default for loan applicants in the process of granting credit. Like any statistical model, scoring is built based on historical data to help predict the creditworthiness of applicants. The method has the drawback of not estimating the probability of default for refused applicants which means that the results are biased when the model is built on only the accepted data set. Reject inference techniques attempt to solve the problem of selection bias by estimating the risk of default of the rejected applicants. First, we present some reject inference techniques like re-weighting, iterative reclassification and parceling, then, techniques that belong to semi-supervised learning mixing labelled and unlabelled data are considered. These techniques are related to mixed classification, adaboost algorithm and gentle adaboost algorithm. ROC analysis is used to compare the performances of different techniques.