Docetaxel-based chemotherapy, as the first-line treatment for metastatic castration-resistant prostate cancer (mCRPC), has succeeded in helping quite a number of patients to improve quality of life and prolong survival time. However, almost half of mCRPC patients are not sensitive to docetaxel chemotherapy initially. This study aimed to establish models to predict sensitivity to docetaxel chemotherapy in patients with mCRPC by using serum surface-enhanced Raman spectroscopy (SERS). A total of 32 mCPRC patients who underwent docetaxel chemotherapy at our center from July 2016 to March 2018 were included in this study. Patients were dichotomized in prostate-specific antigen (PSA) response group (n = 17) versus PSA failure group (n = 15) according to the response to docetaxel. In total 64 matched spectra from 32 mCRPC patients were obtained by using SERS of serum at baseline (q0) and after 1 cycle of docetaxel chemotherapy (q1). Comparing Raman peaks of serum samples at baseline (q0) between two groups, significant differences revealed at the peaks of 638, 810, 890 (p < 0.05) and 1136 cm1 (p < 0.01). The prediction models of peak 1363 cm^-1 and principal component analysis and linear discriminant analysis (PCA–LDA) based on Raman data were established, respectively. The sensitivity and specificity of the prediction models were 71%, 80% and 69%, 78% through the way of leave-one-out cross-validation. According to the results of five-cross-validation, the PCA–LDA model revealed an accuracy of 0.73 and AUC of 0.83.