Power spectral density (PSD) analysis is a widely used spectral method for fatigue assessment of flexible structures due to its computational efficiency and statistical reliability. However, existing approaches often rely on arbitrarily selected PSD datasets that fail to represent the true variability of structural response shapes. This study proposes a systematic methodology for generating a minimal yet representative set of PSDs tailored for vibration fatigue analysis. The method models PSDs as simple rectangular bimodal shapes, defined by three independent parameters linked to classical bandwidth indices. This compact representation prioritizes equivalence in fatigue damage rather than frequency content. Validation against finite element model-derived stress PSDs shows that the bimodal approximation accurately reproduces cycle amplitude distributions and damage potentials, matching established spectral methods. The proposed framework enhances the robustness of representative load case definitions and supports the development and validation of fatigue assessment tools under realistic conditions.