α-Fetoprotein (AFP) is an important tumor biomarker. In particular, the overexpression of AFP-L3 is associated with hepatocellular carcinoma (HCC). Accordingly, several hospitals have begun to employ the ratio of AFP-L3 to the total AFP level (AFP-L3%) as new diagnostic evidence for HCC owing to its high diagnostic accuracy. However, current methods of detection for AFP and AFP-L3 are time-consuming, require multiple samples, and lack in sensitivity and specificity. Herein, we present a novel concept for the early diagnosis of HCC based on the combination of Raman frequency shift and intensity change, and developed surface-enhanced Raman scattering (SERS)-based immunochips via AFP-L3%. In the first step of the study, the frequency shift of 4-mercaptobenzoic acid (MBA) was applied for the quantitative determination of total AFP based on the AFP and anti-AFP interaction on MBA-modified silver chips. 5,5-Dithiobis(succinimidyl-2-nitrobenzoate) (DSNB)-modified immunogold was then incorporated with AFP-L3 antibodies for sandwich immunoreaction on the chips. As a result, we found that a typical Raman band intensity of DSNB presented an exponential linear relationship with the concentration of AFP-L3. Thus, the AFP-L3% can be calculated according to the concentrations of AFP-L3 and total AFP. The most important advantage of the proposed method is the combination of AFP-L3% and frequency shifts of SERS, which exhibits excellent reproducibility and high accuracy, and significantly simplifies the conventional detection procedure of AFP-L3%. Application of the proposed method with the serum of patients with HCC demonstrated its great potential in early liver cancer diagnosis.