In order to achieve higher spectral accuracy and chromaticity accuracy, a residual compensation of weighted spectral dimension reduction method based on modified LMS was proposed in this paper. A homemade color test target composed of 7986 color patches was produced using Matlab. The first 5986 color patches were used as the training samples, while the other 2000 color patches and Munsell color swatches were used as the test samples. The spectral data of the training samples were measured and then the high dimensional spectral data′ s dimensionality was reduced by the residual compensation of weighted spectral dimension reduction method based on modified LMS, and the dimensionality reduction effect was compared with those of the principal component analysis, LabPQR nonlinear dimension reduction method, WSPCAplus nonlinear dimension reduction method and modified matrix R dimensionality reduction method. Experimental results showed that the reconstruction spectral data through the residual compensation of weighted spectral dimension reduction method based on Modified LMS had higher spectral accuracy and chromaticity accuracy, and it still had a stable color accuracy in the case of different observation view angles. In conclusion, this paper proposed a residual compensation of weighted spectral dimension reduction method based on modified LMS, the spectral data reduced by this method had a good spectral accuracy and chromaticity, as well as a good practicability.