For building energy performance prediction and assessment, two approaches are widely used: simplified calculation (e.g. ISO 13790) and dynamic simulation (e.g. EnergyPlus). The ISO 13790 standard uses simple algebraic equations, while dynamic simulation is focused on transient behavior of systems and buildings. This study aims to compare the aforementioned two approaches under uncertainty. For this study, an office building was selected and modeled using ISO 13790 and EnergyPlus 6.0 with the assumptions as close to each other as possible. The sensitivity analysis was employed to identify unknown inputs that have important bearing on the simulation output. Then, Latin Hypercube Sampling (LHS) method, one of the Monte Carlo techniques, was employed for uncertainty propagation. It was found that the two approaches (ISO 13790 vs. EnergyPlus) have different population means from each other and require careful calibration for utilization factors in the simplified approach (reference numerical parameter, reference time constant for both heating and cooling). To calibrate the unknown parameters in the simplified approach, Bayesian calibration was applied in this study. Bayesian calibration is useful to obtain the posterior distribution for the unobserved quantities based on the presumed prior distribution. It is concluded that the simplified approach, when stochastically calibrated, becomes surprisingly similar to the dynamic simulation. (C) 2013 Elsevier B.V. All rights reserved.