Large cellular box beams, 8000 mm long, were fatigue tested in four-point bending. A total of ten tests were performed in a test matrix which included several variations in load ratio. The nominal stress range varied from 89 to 219 MPa. The box beams were fabricated from HSLA-80 steel. Crack growth rates in the base metal were obtained from CT specimens of various heats, orientations and thicknesses. In every test, a through-thickness crack developed at an intentionally poor weld detail after less than 30% of the total fatigue life. Nearly fixed crack growth rates were observed for a region of cracking, due to web restraints, redundancy aspects of the box beam design, and welding residual stresses. Cracks 1500 mm or more in total length still grew in a stable manner. The observed crack growth rates were related to macroscopic beach marking of the fatigue crack surface as well as microscopic striations. Correlation of the observed crack growth rates to the square of stress range was relatively good. The cracking history is decomposed into four separate stages. Using the concept of effective stress intensity factor range, crack growth rates predictions for each stage are presented.