Abstract: As an important cellulose raw material in textile industry in China, ramie needs a fast and accurate analysis method to determine its chemical composition. In this research, calibration models were established using near-infrared (NIR) spectroscopy assisted with partial least square (PLS) to predict the main chemical compositions of ramie based on the prior work using Acousto-optic tunable filter (AOTF) near infrared spectrograph, and the thickness of samples affecting calibration model was also discussed. Results show that the average relative error (ARE) between the predict value and the wet chemistry measured value of cellulose and gum content were 1.11% and 4.54%, which were low enough for prediction. Meanwhile, the thicker of sample, the more accurate for NIR model to prediction.