AbstractSeveral alternatives exist to compute seismic fragility curves. This study takes advantage of the large pool of site‐specific CyberShake simulated ground motions (GMs) to investigate the sensitivity of fragility curves to multiple analysis parameters: the analysis method (Multiple Stripe Analysis (MSA), Cloud Analysis (CA), or Incremental Dynamic Analysis), the number and intensity measure distribution of the GMs used, the GM selection method, and the amount of scaling. To this end, the fragility curve of a two‐dimensional steel frame is calculated for every analysis variation at the life safety limit state. By varying one parameter at a time, we can separate the effects of the various parameters from one another. We also assess the effect of the analysis parameters on the mean annual rate of exceedance of life safety. We find that if GMs are selected adequately for each method, different analysis methods can lead to consistent mean annual rates of exceedance despite some differences in their fragility curves. Generally, MSA and the proposed CA that models the increase of response variability with ground motion intensity best match empirical fragility points. The number of GMs affects the results for all analysis methods, while the intensity distribution of GMs affects results differently in different methods. When GMs are required to match earthquake scenario parameters in addition to intensity, more conservative fragility curves are obtained. Finally, fragility curves are sensitive to excessive scaling. This study provides important insights for performance‐based earthquake engineering.