Abstract: Exploring the effect of defects on the fatigue performance of welded joints is an important content in controlling the quality of welding process. This paper introduces the defect effects, based on cumulative life-critical damage to establish fatigue reliability analysis model of misaligned load-carrying cruciform welded joints. First, the finite element models were established based on the geometrical and defect characteristics and cyclic loading conditions of experimental specimens. Then the fatigue performance of local parameters were explored based on the average strain energy density method (SED) and hot spot stress method (HS). Secondly, a reliability analysis model was established by combining BP neural network and Miner's linear cumulative damage theory to quantitatively analyze the effects of cycle loading, coefficient of variation on fatigue reliability and fatigue damage. The results show the axial and angular misalignments and their probability distribution parameters significantly affect the fatigue life distribution and reliability probability of welded joints. The reliability model provides a reference method for fatigue life design to monitoring and maintenance engineering weldments.