Numerous models to predict the shear strength of reinforced concrete structural walls have been proposed in the literature. Evaluation of the predictive performance of new models relative to existing models is often challenging because the models were created with different levels of complexity and calibrated using different databases. More complex models are expected to have less variance than simpler models, and target performance metrics for models of different complexity do not exist. In addition, a common, comprehensive database should be used to enable direct comparisons between different models. To address these issues, the present study applies statistical and machine-learning approaches to propose a five-step framework to establish target performance metrics for models with different levels of complexity. Application of the framework is demonstrated by addressing the problem of estimating wall shear strength using a comprehensive database of 340 shear-controlled wall tests.