The use of nanozymes has become a promising auxiliary approach to "turn on" surface-enhanced Raman scattering (SERS) signals for the label-free detection of disease markers. Nevertheless, there are still major challenges to develop bifunctional nanomaterials with both excellent enzyme-like activity and high SERS performance. To this end, a novel Z-scheme MoO3-x/CuS heterojunction was first constructed as a powerful "two-in-one" substrate, which can not only catalyze leucomalachite green (LMG) to SERS-active malachite green (MG) but also serve as an efficient substrate to amplify the SERS signal of catalysate. Due to the strong interfacial coupling effect between the MoO3-x and CuS nanomaterial, which promoted the separation and transport of carriers in the heterojunction, the MoO3-x/CuS heterojunction showed higher peroxidase-like activity compared to individual components and the previously reported heterojunction nanozymes. Inspired by these results, a sandwich-type SERS immunoassay for the detection of the cerebral infarction biomarker S100 calcium-binding protein (S100B) was proposed based on the output signal of MG at 1620 cm-1. Furthermore, introducing the antifouling material chitosan on the surface of the MoO3-x/CuS heterojunction can effectively resist nonspecific protein adsorption and significantly improve the detection accuracy of the immunoassay. Therefore, the SERS immunoassay based on the MoO3-x/CuS heterojunction realized highly sensitive and selective detection of S100B in the concentration range of 0.001 to 100 ng/mL, with a low limit of detection of 0.47 pg/mL. The developed method has been successfully used for the accurate detection of S100B in clinical serum. The results showed that the level of S100B in the serum of cerebral infarction patients can be distinguished from those of healthy individuals and intracranial tumor patient controls. In addition, the acquired values of S100B in the serum of cerebral infarction patients based this strategy were well consistent with the results of electrochemiluminescence (ECL) detection with a relative error of less than ±7.3. It is expected that this work may open up a paradigm for improving detection sensitivity and accuracy for the early diagnosis and treatment monitoring of cerebral infarction in the clinic.