Due to advancements in computational tools, performing a full nonlinear response history analysis of structures at the system level using detailed numerical simulations of components has become less demanding, even on standard personal computers. However, the modelling approaches need to be calibrated and verified against experimental measurements to ensure the accuracy of the system-level predictions. In this study, an advanced numerical simulation method is used to predict the 3D nonlinear dynamic response of a four-story reinforced concrete building tested at E-Defense shake table. The structure is modelled in the finite element analysis programme DIANA, using a previously developed and validated approach to predict the failure modes of doubly reinforced walls with confined boundary regions. Curved shell elements are used to model the walls and slabs, and the reinforcement is modelled using embedded bar elements. The frame elements are modelled using beam elements with the effect of confinement on the concrete behaviour represented in the material constitutive model. The numerical versus experimental response comparison is conducted at both local and global levels. For local levels, the base shear versus top displacement curves, strain gradients as well as the crack pattern predicted by the numerical model are compared with the test measurements. The inter-story drift response history is considered as the metric for the global level assessment. The effects of different modelling parameters on the accuracy of predictions are also investigated. ●Development of a detailed yet feasible finite element modelling approach for nonlinear response prediction of RC buildings.●Verification of the model using full-scale shake table test of a four-story RC wall building.●Sensitivity analysis of the nonlinear response prediction to different modelling parameters.●Evaluation of the prediction accuracy with respect to local and global response parameters.●Recommendations regarding an optimum modelling approach that would result in fairly accurate predictions with reduced simulation complexity.