Abstract: To achieve the rapid and nondestructive detection of main active ingredients (reducing sugars, total flavonoids, chlorogenic acid, rutin, and cembratriene-diol) in tobacco flowers, an analytical model for these five components was established via near-infrared (NIR) spectroscopy. A total of 196 tobacco flower samples with different origins and varieties across China were used as research materials. Model performance was further improved by screening spectral preprocessing methods, optimizing spectral ranges, and adjusting the number of principal components. The optimal model for each constituent was obtained using internal cross-validation. Results showed that for the calibration models of the five active ingredients, the correlation coefficients of the calibration set (R) ranged from 0.957 to 0.986, with root mean square errors of calibration (RMSEC) between 0.025 and 0.995; the correlation coefficients of the prediction set (R) varied from 0.901 to 0.979, with root mean square errors of prediction (RMSEP) ranging from 0.0303 to 1.3500. External validation results demonstrated that the relative standard deviations (RSDs) of all five models were less than 10%. Additionally, t-tests revealed no significant difference between the values predicted by NIR spectroscopy and those obtained via traditional chemical analysis. This confirms that the proposed NIR spectroscopy method exhibits high accuracy and reliability, and enables the rapid and nondestructive detection of the five main active ingredients in tobacco flowers, which thus provides technical support for the efficient development and utilization of tobacco flower resources.