On the basis of the principal component analysis algorithm, a residual compensation of weighted spectral dimension reduction model based on LMS was proposed in this paper. The basic framework for using the LMS weight function for weighting the original spectra and using the residual spectra for model compensation was introduced. The Munsell color cards were chosen as the training samples, while the SG color cards and multi-spectra images were chosen as the test samples. The RCwPCA proposed in this paper was compared with PCA in compressing and reconstructing the training and test samples. Experimental results showed that reconstruction by RCwPCA could reach higher chromaticity accuracy under different dimensions. This algorithm effectively improved the chromaticity accuracy of PCA and kept higher chromaticity stability under the condition of variable light sources. The RCwPCA dimension reduction model which used LMS weighting and residual spectra compensation was a spectral dimension reduction model with high precision.