Moisture content is considered as an important evaluation index for beef quality. This study focused on development of nondestructive rapid detection system device for assessing beef quality based on VIS/NIR spectroscopy. Working principal and process of the system, hardware composition, and software were introduced. Hardware of detecting system included two spectrometers, handheld probe, control instrument, bifurcated optical fiber, power unit and container. Two spectrometers in the spectral region of 400~960nm and 900~2600nm were coupled together for spectral acquisition. The wavelength range of visible and near infrared spectroscopy was covered and the characteristic wavelengths in VIS/NIR band of moisture, protein, fat and other main components of beef were detected by the two spectrometers. Another important part was probe. The handheld probe was designed with two optical channels in order to reduce the effect of light source on fiber probe. One channel was designed with angle of 45° for light source and the other one was designed to be vertical for the fiber probe. Inner surface of the two channels was covered by white barium sulfate to make the light distribution even. The handheld probe was designed to ensure the same distance between the end of fiber probe and the surface of sample. Experiment showed that different samples could be detected by adjusting distance between the end of fiber probe and the bottom surface of handheld probe. One end of the bifurcated fiber was connected with handheld probe and the other two were connected with two spectrometers. Software of spectral data collecting and rapid detection was developed by using VC++, and can be run in Windows environment. Main modules of the software included parameter setting module, sample information management module, trigger control module, spectral information acquisition module, quality evaluation module and results display and storage module. The system could collect spectral data, process the data, detect the quality of sample and display the results. First, the system was used to acquire optical data from 57 beef samples of M. longissimus dorsi, and build the prediction model of beef moisture content by visible spectra, NIR spectra, full spectra, respectively. The results showed that the prediction model developed by full spectra had the highest accuracy. The correlation coefficient of calibration (Rc) and the correlation coefficient of prediction (Rp) of the prediction model were 0.96 and 0.88, respectively. Then, experiment on the system was done to detect moisture content of beef on the processing line in two beef slaughtering enterprises. In this test, 84 beef ridge samples were extracted for moisture content detection. The detection results from the experiment yielded satisfactory results. Correct rate of non-destructive rapid detection system device testing on moisture content was 92.8%. The result shows that the system can be used for nondestructive rapid detection of beef quality with high accuracy and reliability as well as repeatability.