Near-infrared (NIRS) spectroscopy, coupled with partial least squares regression, was used to predict adenosine and cordycepin concentrations in fruiting bodies of Cordyceps militaris . The fruiting body samples were prepared in four different sample formats, which were intact fruiting bodies, chopped fruiting bodies, dried powder, and dried crude extract. The actual amount of the adenosine and cordycepin concentrations in fresh fruiting bodies was analyzed by high-performance liquid chromatography. Results showed that the prediction models developed from the chopped samples provided excellent accuracy in both parameters with minimal sample preparation. These optimum models provided a coefficient of determination of prediction, standard error of prediction, bias, and residual predictive deviation, which were respectively 0.95, 16.60 mg kg –1 , –8.57 mg kg –1 , and 5.04 for adenosine prediction, and 0.98, 181.56 mg kg –1 , –1.05 mg kg –1 , and 8.9 for cordycepin prediction. The accuracy and performance of the model were determined by ISO12099:2017(E). It was found that these two equations can be considered to be acceptable at a probability level of 95% confidence. The NIRS technique, therefore, has the potential to be an objective method for determining the adenosine and cordycepin concentrations in C. militaris fruiting bodies.