AbstractThe quality of predicted plant‐, soil‐, and animal‐response values from near‐infrared (NIR) reflectance spectra depends on the ability to generate appropriate NIR models. The first step in the development of NIR models is collection of spectral data. Limited work, however, has been reported that compares NIR models for prediction of forage nutritive value when the spectra are obtained from devices with different spectral ranges and resolutions. The objectives of this study were to (a) develop and evaluate NIR spectroscopy models using a benchtop‐type (FOSS) and two handheld NIR devices (microPHAZIR and DLP NIRscan Nano EVM) to predict crude protein (CP), acid detergent fiber (ADF), amylase and sodium sulfite‐treated neutral detergent fiber (aNDF), and in vitro true dry matter digestibility (IVTD) of dried ground forage grass samples and (b) compare predictions among the three NIR devices. Switchgrass (Panicum virgatum L.) and bermudagrass [Cynodon dactylon (L.) Pers] hay samples were scanned with the NIR devices and analyzed with wet chemistry for development of NIR prediction models. Among devices, the r2 of validation values for aNDF models ranged from .81 to .87; all other r2 values were >.86 and as high as .98 with standard error of prediction (SEP; g kg−1) ranging from 8.1 to 11.5 for CP, 19.1 to 23.8 for aNDF, 14.2 to 20.0 for ADF, and 26.8 to 49.9 for IVTD. The FOSS benchtop NIR prediction models consistently had the highest r2 and lowest SEP values; however, the predictive power for both handheld devices was similar to the benchtop‐type device.