A spectrum signals detection method has been designed for surface enhanced Raman scattering (SERS) under high fluorescence and background noise. The components of the fluorescence and background noise in SERS spectrum signal were analyzed first. Then they were evaluated by some models, such as polynomial model and AR model. By analysing the difference between evaluation result and original data, the Raman spectrum peaks can be detected. After being used in Rhodamine 6G, prostate-specific antigen and PH sensing experiment, the method was proved to be of high performance and practicability for SERS signal processing.