We set out two generic principles for selective editing, namely the minimization of interactive editing resources and data quality assurance. These principles are translated into a generic optimization problem with two versions. On the one hand, if no cross-sectional information is used in the selection of units, we derive a stochastic optimization problem. On the other hand, if that information is used, we arrive at a combinatorial optimization problem. These problems are substantiated by constructing a so-called observation-prediction model, that is, a multivariate statistical model for the nonsampling measurement errors assisted by an auxiliary model to make predictions. The restrictions of these problems basically set upper bounds upon the modelled measurement errors entering the survey estimators. The bounds are chosen by subject-matter knowledge. Furthermore, we propose a selection efficiency measure to assess any selective editing technique and make a comparison between this approach and some score functions. Special attention is paid to the relationship of this approach with the editing fieldwork conditions, arising issues such as the selection versus the prioritization of units and the connection between the selective and macro editing techniques. This approach neatly links the selection and prioritization of sampling units for editing (micro approach) with considerations upon the survey estimators themselves (macro approach).