Extreme loading scenarios such as earthquakes, exceptional wind gusts or tides are characterized by very high amplitude and multidirectional cyclic loads applied to a structure. Ultra Low Cycle Fatigue (ULCF) arises in such cases and can be summarily defined as the deterioration of material properties due to repetitive loading at large amplitudes. In the case of welded steel joints, failure is generally reached in just a few number of cycles - typically less than ten or twenty. The importance of a comprehensive understanding of this phenomenon lies in the balance between safety or allowable damages and an economical design. For rare events with very high demands there is a need to use the material's properties to its fullest in order to obtain the most rational design. A common example of this is the use of material ductility to absorb the energy of earthquake loading in design standards. The goal of this thesis is to provide the engineer with both the understanding of the physical phenomenon involved in ULCF and the tools to design a structural component with flexibility and sufficient accuracy. Failure of metals in the presence of high plastic strains is commonly observed to be of ductile nature. Micromechanical models based on homogenization theory are typically used to predict this type of fracture, because they attempt to capture the fundamental mechanisms involved in void growth to coalescence. In this thesis, two micromechanical models (Gologanu-Leblond-Devaux for void growth and Torki-Benzerga-Leblond for void coalescence), developed for monotonic loading, are studied in the context of large amplitude cyclic loadings and their predictions are compared with experimental results obtained on small scale specimens for a high strength steel - S770QL. Component scale specimens, namely welded tube to plate specimens of the same steel type, are tested in bending and torsion. Digital image correlation measurements of weld toe strains for over 60 tests are presented and used to recommend a design procedure using local strains in finite element modeling, with a Manson-Coffin type of approach