The present paper presents the necessary crack growth statistics and suggests stochastic models for a reliability analysis of the fatigue fracture of welded steel plate joints. The reliability levels are derived from extensive testing with fillet-welded joints for which the entire crack growth history has been measured, not only the final fatigue life. The statistics for the time to reach given crack depths are determined. Fracture-mechanics-derived crack growth curves are fitted to the measured experimental curves and the best fit defines the growth parameters involved for each test specimen. The derived statistics and distribution function for these parameters are used as variables in a Monte Carlo simulation (MCS). In addition a Markov model is developed as an alternative stochastic model. It is a Markov chain for which the discrete damage states are related to chosen crack depths in the material. This model works directly with the experimental time statistics. It is a “stochastic bulk approach” not involving any random variables or fracture mechanics modeling. Both models are fitted to the data base and scaled to in-service conditions. Both methods are compared and discussed. The aim is to provide data for the variables used in a MCS and to develop a Markov chain for fast reliability calculation, especially when predicting the most likely influence of numerous future inspections.