Fatigue behavior of boxing welded joints of JIS A 5083 P-O Al-Mg alloy was examined. For a specimen, a 20 mm thick stiffener was attached to a 20 mm thick main plate by a MIG boxing welding. Previous to fatigue tests, residual stress measurement, a static loading test and an elastic finite element analysis (FEA) were carried out. The residual stress in the fatigue stress direction was 135 MPa at a distance of 5 mm from a boxing weld toe, and the FEA gave a fairly good estimation for the elastic strain distribution near the boxing weld toe. Then fatigue tests were performed under both constant amplitude and random loads by 3-point bending. For a load wave in the random loading fatigue tests, a direct current component, a zero-mean narrow band random process, and a high frequency component were variously and selectively combined and superimposed. As a result, a reference stress 65, which was determined as the stress at a distance of 5 mm from a boxing weld toe, was proved effective in reducing the amount of scattering in the fatigue test results. And it is also found that the crack initiation lives determined by the 5% drop method largely scattered because of the determination error brought by the 5% drop method itself, showing to be unsuitable for putting the fatigue test results in order as in a S-N diagram. With regard to the effects of high frequency components, it was found that a simple handling of fatigue data using the RMS value of stress is not available while arrangement using a representative frequency is effective, and that neither the range count method nor the peak to peak count method are suitable to those superimposed wide band stress waves. Finally the life estimation procedure proposed in the former report, which comprises a 2-dimensional rainflow method, the modified Goodman's relation and the modified Miner rule, was shown to be valid also in this study.