This study collected the near infrared (NIR) hyperspectral images of 150 astringent persimmons, with the spectra are in 900~1700 nm. Monte Carlo-uninformative variable elimination (MC-UVE) algorithm and successive projections algorithm (SPA) were adopted to the optimization of wavelengths obtained from the region of interest (ROI). Eight wavelengths were selected by MC-UCE-SPA. These feature wavelengths were 924.69, 928.05, 1 112.72, 1 270.91, 1 365.3, 1 402.42, 1 453.06 and 1 547.69 nm, respectively. The spectral reflectance of the 8 feature wavelengths were applied to establish the detective model for the soluble solid content (SSC) of persimmon by partial least squares regression (PLSR) method. The correlation coefficient and root mean square error of prediction set are r_pre=0.942, RMSEP=1.009 °Brix. The results indicated that MC-UVE-SPA could effectively extract the characteristic information related to the SSC and develop a better predictive model with fewer wavelengths. This work can provide technical support and research basis for the nondestructive detection, grading and processing equipment for persimmon quality.