Near infrared (NIR) spectroscopy was employed to perform a quantitative analysis of gentiopicroside, the bioactive component of the medicinal plant Gentiana scabra Bunge. Modified partial least squares regression (MPLSR) and stepwise multiple linear regression (SMLR) calibration models were built using 94 plant tissue culture samples and 136 grown plant samples, respectively, over the full wavelength range (400–2498 nm) and the silicon charge-coupled-device (CCD) sensing band (400–1098 nm). For tissue culture, the smoothing, first-derivative MPLSR model can produce the best effect [calibration set (Rc) = 0.868, standard error of calibration (SEC) = 0.606%, standard error of validation (SEV) = 0.862%] in the wavelength ranges of 900–1000, 1200–1300, and 1600–1700 nm. By contrast, for grown plant samples, the smoothing, second-derivative MPLSR model can produce the best effect (Rc = 0.944, SEC = 0.502%, SEV = 0.685%) in the wavelength ranges of 400–500, 1100–1200, 1600–1800, and 2200–2300 nm. With the silicon CCD sensing band, the smoothing, second-derivative, four-wavelength (670, 786, 474, and 826 nm) SMLR model showed best predictability (Rc = 0.860, SEC = 0.775%, SEV = 0.848%). This study successfully built spectral calibration models for determining gentiopicroside content at different growth stages of G. scabra Bunge. The specific wavelengths selected within the silicon CCD sensing band can be used in combination with multispectral imaging as a powerful tool for monitoring or inspecting the quality of G. scabra Bunge during cultivation.