It is crucial to obtain the internal hardness distribution in polycrystalline materials to evaluate the mechanical performance of components and monitor their service life. Current methods, however, fail to meet the non-destructive evaluation needs for materials with hardness gradient distributions. This paper, based on the principle of grain boundary scattering of ultrasound in polycrystalline materials, combined with the Transverse-to-Transverse Singly-Scattered Response (T-T SSR) theory, proposes an ultrasonic SSR model adapted to hardness gradient distributions. The model elucidates the influence of hardness gradient variations and grain dispersion on ultrasonic scattering. Using DREAM.3D, seven different-scale polycrystalline volumes were constructed to assess the relevance of volume-weighted average grain size and spatial correlation of hardness gradient materials. Finally, induction quenching was applied to 40Cr to induce a gradient hardness distribution internally, followed by ultrasonic backscatter experiments. The results indicate that the theoretical model and the spatial variance of measured signals align well over a relatively long time window. For the specimen with minor curvature, the theoretical hardness distribution obtained by the model is accurate, with an average error of 2.55 % compared to destructive testing data. However, the results for the larger curvature reveal limitations in the model.