Abstract | Following recent guidelines (e.g. FEMA-350) seismic performance uncertainty is an essential ingredient for performance-based earthquake engineering. Uncertainty refers to both aleatory uncertainty, raised by the random record-to-record variability, and also to epistemic uncertainty primarily introduced by modeling assumptions or errors. A methodology for the performance-based estimation of the dispersion introduced by parameter uncertainties is developed. The methodology proposed provides an inexpensive alternative to the use of tabulated values, or to performing a series of time-consuming nonlinear response history analyses to obtain parameter uncertainty. As a testbed, the well-known 9-storey LA9 2D steel frame is employed using beam-hinges with uncertain backbone properties. The monotonic backbone is fully described by six parameters, which are considered as random variables with given mean and standard deviation. Using point-estimate methods, first-order-second-moment techniques and latin hypercube sampling with Monte Carlo simulation, the pushover curve is shown to be a powerful tool that can help accurately estimating the uncertainty in the seismic capacity. Coupled with SPO2IDA, a powerful R-μ-T relationship, such estimates can be applied at the level of the results of nonlinear dynamic analysis, allowing the evaluation of seismic capacity uncertainty even close to global dynamic instability.