Strategies for developing quantitative structure-affinity relationships (QSAfR) for the prediction of break-through performance of 31 chlorinated hydrocarbons on activated carbon have been studied. Two different approaches for the selection of a limited set of compounds for modelling were evaluated through the predictive power of the resulting QSAfR models. When the model was based on a training-set selected without a rational strategy, the developed QSAfR model showed poor predictive performance. Accordingly, such models have a limited capability to produce information concerning the important adsorbate related parameters influencing adsorption. By using a strategy where multivariate data analytical techniques are used in conjunction with statistical experimental design to select a balanced set of compounds for break-through performance evaluation, it was possible to develop QSAfR models with high predictive capability.