A total of 292 air-dried soil samples were used from Heilongjiang Province, and the hyper-spectral reflectance data were also measured in the laboratory. Meanwhile, color components of all soil samples were calculated in the CIE XYZ color space according to the three-stimulus method to predict and validate soil color. Then, Landsat-8 OLI original reflectance data of each soil sample site were extracted to calculate the normalized difference vegetation index, normalized difference water index, normalized difference moisture index and normalized difference impervious surface index. Based on these remote sensing indexes, the threshold value of modeling spectra screening is proposed. Combined with the extracted remote sensing spectra, the partial least squares regression model was used to predict soil color components. The results showed that for the color components of CIE X, Y and Z, the validation R² value were 0.76, 0.76 and 0.69, and the RPD value were 1.74, 1.76 and 1.68, indicating that the model established by partial least squares can be used to predict soil color. The prediction of different land use types showed that cultivated land soil color was better in prediction than those of forest and grassland. Soil color was also predicted under different organic carbon contents, showing the prediction was better in lower organic carbon content.