The seismic performance tests of a full-scale five-story passively controlled steel building were conducted on the E-Defense shaking table in Japan in March 2009. Before the tests, a blind prediction contest was held to allow researchers and practitioners from all over the world to construct analytical models and predict the dynamic responses of the steel frame specimen equipped with buckling-restrained braces (BRBs) or viscous dampers (VDs). This paper presents the details of two refined prediction models made and results obtained before the tests. When the proposed analytical modeling techniques are adopted as in the two refined prediction models, the overall prediction accuracy is about 90%. Sensitivity studies conducted after the tests are also presented in this paper. The effects of varying each modeling feature on the response simulation accuracy have been investigated. The analytical results suggest that considering concrete full-composite actions for beam members could improve prediction accuracy by about 20% against using the simplified bare steel beam model. Adopting refined BRB stiffness computed from incorporating finite-element gusset stiffness only improves the overall prediction accuracy by 0.9%. Considering the BRB dynamic loading test results for analytical BRB strength reduces the error by 1.9%. For the VD frame, incorporating the brace and VD stiffness could improve the overall prediction accuracy by about 15%. Copyright © 2012 John Wiley & Sons, Ltd.