In this paper, Raman spectroscopy of serum was detected and analyzed by multivariate analysis and classification methods to identify its diagnosis potential for liver diseases. Serum taken from 81 volunteers (31 healthy, 27 liver cancer and 23 liver cirrhosis) was measured. Principal component analysis (PCA) was utilized on the first order derivative spectra of the original Raman spectroscopy. Linear discriminant analysis (LDA) and artificial neural networks (ANN) were then used separately on the selected PCs obtained from PCA. The diagnosing accuracy of PCA-LDA is 85%, and is 80% for PCA-ANN. In conclusion, this preliminary study demonstrated that Raman spectroscopy of serum can be used in the screening of liver cancer.