Before bolts were invented, hot-riveting was employed as technological process to produce built-up structural members of steel structures. Nowadays, these structures being still in service and susceptible to fatigue failure, they need to be assessed for their remaining service life. To estimate the fatigue life distribution of riveted connections is, therefore, a topic of relevance. A system reliability model for estimating the probabilistic fatigue life of riveted shear connections is proposed in this paper. Similarly to other models available in the literature, the failure of the connection is modelled by evaluating potential failure at identified critical locations. In addition to these existing models, the proposed formulation is able to consider: (1) the dependency between failure at different critical locations by updating the state of stress given that a failure has occurred and evaluate the residual fatigue life considering the cumulated damage, and (2) the fail-safe behaviour of the connection by using system reliability. Thus, it is possible to evaluate the effect of the position and the number of the rivets on the fatigue life of the connection. The stress and strain field is obtained using the finite element method and the fatigue life estimation is performed using the strain-life approach. The fatigue resistance of the plain material, the rivet clamping force, and the friction coefficient are considered as stochastic quantities. The constant amplitude fatigue life predicted by the reliability model is compared with three datasets of riveted double shear connection having different geometries. In addition, several analyses are performed: (1) a sensitivity study to determine the relationship between the input parameters and the fatigue life, (2) a comparison with other reliability models, and (3) a comparison with a similar connection having a different number of rivets.