Wheat protein content is an important indicator often employed in wheat sale price. Spectral indexes and concurrent winter wheat protein content (WWPC) samples were obtained across three years. Data from 2008/2009 and 2009/2010 were utilized to build the new ratio indexes and product indexes, and then selected grey relational method and partial least squares method were used to improve the estimation accuracy of WWPC, data from 2011/2012 was utilized to validate model. The results showed that the correlation coefficients between ratio indexes and WWPC were better than that between single indexes and WWPC. The r of single indexes and ratio indexes were 0.726 and 0.751, respectively, and the product indexes were used to improve the parts of single indexes. The estimation accuracy of WWPC was improved by using GRA-PLS, the determination coefficients (R2) of single indexes, ratio indexes and product indexes were 0.537, 0.631 and 0.521, respectively, and corresponding root mean square errors (RMSE) were 0.665%, 0.564% and 0.574%, respectively. The results indicated that it was feasible to estimate WWPC by building new ratio indexes and product indexes, and then applying the GRA-PLS.