With the non-ionizing, non-invasive, high penetration, high resolution and spectral fingerprinting features of terahertz (THz) wave, THz spectroscopy has great potential for the qualitative and quantitative identification of key substances in biomedical field, such as the early diagnosis of cancer, the accurate boundary determination of pathological tissue and non-destructive detection of superficial tissue. However, biological samples usually contain various of substances (such as water, proteins, fat and fiber), resulting in the signal-to-noise ratio (SNR) for the absorption peaks of target substances are very small and then the target substances are hard to be identified. Here, we present recent works for the SNR improvement of THz signal. These works include the usage of attenuated total reflection (ATR) spectroscopy, the fabrication of sample-sensitive metamaterials, the utilization of different agents (including contrast agents, optical clearing agents and aptamers), the application of reconstruction algorithms and the optimization of THz spectroscopy system. These methods have been proven to be effective theoretically, but only few of them have been applied into actual usage. We also analyze the reasons and summarize the advantages and disadvantages of each method. At last, we present the prospective application of THz spectroscopy in biomedical field.