Noni powders with the moisture content of 15.33% were prepared from fresh noni fruit by spray drying, then vacuum freeze drying to obtain samples with moisture content (0.44%～9.00%) by different dying time respectively. Then the model between IR and water content of all samples with partial least squares regression (PLSR) was established. The results showed that IR absorption peak of water appeared about 1 640 cm^-1, and the absorption peak was influenced by the freeze drying time. PLSR model with moisture content in range of 0.44%～9.00% in the noni fruit powder was also built, and the square of determination coefficient for prediction is 0.91, with the interactive authentication root mean square error (RMSECV) 0.62. However, the square of determination coefficient for prediction was 0.98, with the interactive authentication root mean square error (RMSECV) 0.26 in the model for moisture content in the noni fruit powder from 5.00% to 9.00%. It showed that the predicted value of moisture content was close to the measured value. The infrared spectroscopy could be used to detect the content of the water (0.44%～9.00%) in the noni fruit with high accuracy in the moisture content range from 5.00% to 9.00%.