Under the influence of random loads such as those from vehicles, the welded details in steel bridges tend to gradually develop into macrocracks, typically originating from manufacturing defects or stress concentration points. As the service life of the steel bridge increases, the continuous generation and propagation of fatigue cracks lead to a decline in the bridge's service performance, potentially resulting in catastrophic failures such as bridge collapse. Accurately analyzing the fatigue crack growth patterns at welded joints is a prerequisite for reliably assessing the structural service performance. The analysis of fatigue crack propagation depends on both the crack growth step size and the cyclic stress–strain response at the welded joint under loading. Considering the complexity of the material properties and stress concentration at the welded joint, which leads to a lack of analytical solutions for the cyclic stress–strain characteristics at the crack tip, numerical analysis was employed to obtain the cyclic stress–strain response at the crack tip of the welded joints. Using the size of the cyclic plastic zone at the crack tip as an adaptive crack growth step size, a model for fatigue crack growth was developed based on the plastic strain energy of the cyclic plastic zone. Fatigue test specimens extracted from the welded toe of cruciform welded joints were tested, providing cyclic stress–strain (Ramberg-Osgood material parameters) and fatigue damage characteristics (Manson-Coffin material parameters) of the material at the welded toe. High-cycle fatigue tests using cruciform welded joint specimens determined a good correlation between the crack propagation characteristic dimensions obtained from the adaptive numerical analysis model and the experimental results. The experimental findings confirmed the applicability of the established adaptive numerical analysis model for studying fatigue crack growth in cruciform welded joints. The parameters introduced in this model are physically meaningful, easy to obtain, and suitable for engineering applications.