When theoretical methods are used to predict the properties of a given system, such as the effects of the substitution of a specific amino acid on the activity or stability of a protein as a whole, the accuracy of the prediction is directly dependent on the validity of the underlying model. A common error, however, is to attempt to improve a basically crude model by performing one aspect of the calculation in a rigorous manner. The accuracy of the model as a whole will remain limited by the crudest approximation or weakest assumption. To demonstrate the principle that nothing can be gained by performing extensive calculations using a basically crude underlying model we compare the predictive power of three models in relation to activity and stability data for 78 triple-site sequence variants of the lambda-repressor protein. This system has recently been analysed in terms of a conceptionally simple, but computationally elaborate model for the prediction of the energy of a protein in which amino acid residues in the core of the protein have been mutated. We show that comparable, if not better agreement with the experimental data can be reached using either of two much simpler models, based on straightforward structural considerations, which do not require elaborate calculations on a computer.