Abstract: It is difficult to obtain the fatigue properties of key engineering structures, small sized parts and welding areas by using low cycle fatigue testing specimens which are given by current standards. In this paper, a new method is proposed to obtain the cyclic constitutive relationship and low cycle fatigue life of materials by using a millimeter-sized specimen. The low cycle fatigue test of TA17 alloy slice is carried out by designing semicircular notched slice specimens, loading fixture and testing scheme. Based on the assumed strain energy separation function, a universal prediction model of cyclic constitutive relationships is proposed. The forward and backward prediction of finite element simulations show that the new model has universal validity for different power-law materials and different dimensions of notched slice specimens. Symmetric strain-controlled variable-amplitude low cycle fatigue tests and multi-level low cycle fatigue tests of TA17 alloy straight round bar specimens and 1.2 mm thickness notched slice specimens are completed. After using the cyclic constitutive relationship of TA17 alloy predicted by the new model to compare with straight round bar specimens, the results show that they are in a good agreement with elastic segments and the large strain segments (0.009 mm/mm~0.011 mm/mm), and that their maximum relative predicting error is less than 9% in elastic-plastic transition segments (0.004 mm/mm~0.009 mm/mm). Adopting the finite element method based on the cyclic constitutive relationships obtained by the new model and straight round bar specimens, two sets of transforming equations are conducted to convent the testing strain amplitude and average stress amplitude of notched slice specimens to axial strain and stress amplitude at the semicircular notched root. Finally, the notched slice specimens' fatigue-life curves of material representative volume element and Manson-Coffin life prediction model via two sets of transforming equations are verified to be identical to the experimental results from straight round bar specimens.