Smart materials such as shape memory alloys have unique material properties that can potentially mitigate earthquake hazards on the built environment. Implementation of shape memory alloy-based devices on building structures should incorporate two key factors: (1) distinct mechanical features of the devices and (2) inherent large uncertainty stemming from material properties, building geometry, and ground motions. This study conducts seismic fragility analyses of steel building frames installed with superelastic shape memory alloy dampers, which enable both factors to be appropriately considered. First, a thermomechanical constitutive model is utilized to capture all essential characteristics of the shape memory alloy damper. Next, a probabilistic seismic analysis framework is developed to obtain the seismic demands of three critical engineering demand parameters (i.e. peak interstory drift ratio, residual drift ratio, and top floor acceleration) of the building when subjected to modeling uncertainty and a large set of realistic ground motion inputs. Nonlinear time history responses and the associated short-time Fourier transform demonstrate the superior control efficiency of the shape memory alloy damper in limiting the building’s residual drift and top floor acceleration. Furthermore, seismic fragilities of the buildings when installed with shape memory alloy dampers are compared with those when equipped with yielding dampers. The study indicates that under different levels of ground motions and various ranges of modeling uncertainty in structural parameters, shape memory alloy damper consistently outperforms the yielding damper in reducing the seismic fragility of the building at both component and system levels.