The unreasonable use of antibiotics in poultry breeding has led to the frequent occurrence of antibiotic residues in poultry products, affecting food safety and posing a threat to human health. Therefore, accurate and rapid detection of antibiotic drug content is of great significance. This study successfully carried out larger and smaller gradient whole content detection of norfloxacin This study successfully carried out larger out based on terahertz spectroscopy. First, in the samples of higher-gradient norfloxacin, 11 concentrations were set in the range of 1%~100% to complete the preparation of high-gradient samples. Then the time spectrum was scanned by the terahertz time spectrum system, the absorption coefficient of the sample was extracted, and the Savitzk-Golay(S-G) binomial fitting filter was used to remove the noise and smooth the spectral data of the sample. It was found that the absorption coefficient of pure norfloxacin had a strong absorption peak at 1.205 THz and a weak absorption peak at 0.816 THz. Finally, stepwise regression and successive projections algorithm (SPA) is used to select variables and combine the characteristic absorption peak to realize the prediction analysis of multiple linear regression modeling. Further, in the study of smaller gradient norfloxacin samples, we set 29 concentration series below 100 μg·mL^-1 (0.01%) to complete the preparation of smaller gradient samples. Then the terahertz time domain spectrum was obtained, the S-G binomial fitting filter was used for data preprocessing, and no significant difference was found in the absorption spectrum of each concentration. Finally, stepwise regression and SPA were used to select variables and combine the characteristic absorption peak to realize multiple linear regression modeling and prediction analysis. The results showed that the multiple linear regression of stepwise regression selection variable in the higher-gradient norfloxacin sample reached the optimal model (R_p=0.962, RMSEP=2.74%), and the accuracy was better than the existing optimal prediction model (R_p=0.867, RMSEP=16.6%). The multiple linear regression model of the stepwise regression selection variable of the small gradient norfloxacin sample is optimal (R_p=0.728, RMSEP=18.79 μg·mL^-1). This method has a certain prediction ability, but its accuracy needs to be improved. In this study, terahertz spectroscopy was used to detect norfloxacin in full concentration, which provided a certain research basis for exploring norfloxacin detection limit and other further studies.