It is known that compressive plastic strain deteriorates the ductility and fracture toughness of steels. So when buckling collapse takes place in structural steel members under the cyclic bending load, cracks may easily initiate and propagate at the concave side (compressive side) of buckling deflection. In this paper, an attempt was made to investigate the crack initiation behavior and quantify the crack initiation strength for structural steel members subjected to cyclic bending load after undergoing buckling collapse. For this purpose, crack initiation tests using strip smooth specimens and fillet welded specimens under the cyclic bending load were performed with JIS SM 490 C steel plate. In these tests the specimens were given various cyclic compressive displacement amplitude after undergoing buckling and number of cycles to crack initiation and fracture were examined. Elastoplastic large deformation FEM analyses were also performed for all loading cases to investigate stress and strain distributions at crack initiation sites. It has been found that : (1) According to the crack initiation tests, for smooth specimens and fillet welded specimens, cracks initiated and propagated from the center of width on the compressive side (concave side). Finally all specimens were fractured due to propagation of these cracks from compressive side. (2) According to the FEM analyses, the largest longitudinal stress and strain were induced at the center of width on the compressive side. With the increase in number of cycles, stress was increased and strain was decreased. (3) Relationships between compressive displacement amplitudes and number of cycles to crack initiation and to fracture can be expressed by linear functions on logarithmic scale. Relationships between longitudinal strain of the crack initiation site at the first compression and number of cycles to crack initiation and to fracture can be also expressed by linear functions, and the relationships for smooth specimens and fillet welded specimens were almost the same.