Rapid determination of moisture content is an important requirement to ensure the production quality of puffs. In this paper, the NIR spectra of 130 modeling samples and 30 validation samples were collected, using IAS Online-S100 Near Infrared Spectroscopy Analyzer. The moisture analysis model of puff was established using spectral pretreatment and partial least square method. The results showed that determination coefficient R2, root mean square error of correction set (RMSEC), root mean square error of interactive verification (RMSECV) and root mean square error of prediction set (RMSEP) of the model are 0.88, 0.49%, 0.55% and 0.57% respectively, when using moving window smoothing method & SNV as spectral pretreatment method and the best principal factors was 9. The prediction error of the model is within ±1.3, and the accuracy meets the needs of the factory.