Standards and guidelines for the fatigue design of riveted connections make use of a stress range-endurance (S-N) curve based on the net section stress range regardless of the number and the position of the rivets. Almost all tests on which S-N curves are based, are performed with a minimum number of rivets. However, the number of rivets in a row is expected to increase the fail-safe behaviour of the connection, whereas the number of rows is supposed to decrease the theoretical stress concentration at the critical locations, and hence these aspects are not considered in the S-N curves. This paper presents a numerical model predicting the fatigue life of riveted connections by performing a system reliability analysis on a double cover plated riveted butt joint. The connection is considered in three geometries, with different number of rivets in a row and different number of rows. The stress state in the connection is evaluated using a finite element model in which the friction coefficient and the clamping force in the rivets are considered in a deterministic manner. The probability of failure is evaluated for the main plate, and fatigue failure is assumed to be originating at the sides of the rivet holes, the critical locations, or hot-spots. The notch stress approach is applied to assess the fatigue life, considered to be a stochastic quantity. Unlike other system reliability models available in the literature, the evaluation of the probability of failure takes into account the stochastic dependence between the failures at each critical location modelled as a parallel system, which means considering the change of the state of stress in the connection when a ligament between two rivets fails. A sensitivity study is performed to evaluate the effect of the pretension in the rivet and the friction coefficient on the fatigue life.