Combining membrane electrophoresis with surface-enhanced Raman scattering (SERS) spectroscopy, the serum proteins were first purified and then mixed with silver nanoparticles to perform SERS spectral analysis. Therefore, the spectral signatures were enhanced to high-fidelity SERS signatures because of the purification procedure of the first step. We used the method to analyze blood plasma samples from nasopharyngeal cancer patients (n = 43) and healthy volunteers (n = 33) for cancer detection. Principle component analysis of the SERS spectra revealed that the data points for the cancer group and the normal group form distinct, completely separated clusters with no overlap. Therefore, the nasopharyngeal cancer group can be unambiguously discriminated from the normal group, i.e., with both diagnostic sensitivity and specificity of 100%. These results are very promising for developing a label-free, noninvasive, and reliable clinical tool for rapid cancer detection and screening. Copyright © 2011 John Wiley & Sons, Ltd.