This work aimed to establish a fast and accurate method to detect glioma by combining surface‐enhanced Raman scattering (SERS) and mathematical analysis. At first, 785‐nm laser was selected as the optimum laser to acquire Raman spectra of human brain tissue. Second, it was verified that Raman data in the range of 1,200–1,600 cm−1 could improve the performance of classifier. Based on the analytical results of 1,200–1,600 cm−1 data, the sensitivity and specificity of partial least square (PLS) analysis and back‐propagation neural network (BPNN) were as high as 100%, whereas the sensitivity and specificity of support vector machine (SVM) were 96% and 100%, respectively. Among them, PLS was more potential in the detection of glioma, because of its lower computational cost compared with SVM and BPNN. Moreover, the correlation between observed Raman peaks and 2‐hydroxyglutarate (2HG; 512, 790, 1,204, 1,302, and 1,463 cm−1) was observed, suggesting 2HG as a potential marker of glioma using its Raman spectroscopic signatures. After all, SERS combining with mathematical analysis could be a promising tool for the accurate detection of glioma.