We contribute to digital analysis by developing two mathematical programming models that can assist auditors in selecting more promising audit samples, using Benford's law. One model identifies the smallest subset of nonconforming records in a dataset, given some predefined conformity criteria, and the other highlights the k most nonconforming records. The models take into account several conformity tests and test statistics simultaneously. The application of the models is illustrated using suggested protocols on a set of simulated data. Finally, the effectiveness of the models in detecting typical data manipulations is assessed under different contamination levels.