To make full use of the hyperspectral data, the strong multi-collinearity in the data is supposed to be taken into account. With this study we evaluated three multivariate regression methods which are principal component regression, partial least square regression (PLSR) and stepwise multiple linear regression. Furthermore, to identify reliable winter wheat biomass predictive models, two different types of spectral transformations (continuum removal, first derivative) were combined with the three regression methods, respectively. Amongst these combinations, the respective combination of three regression methods and continuum removal got the highest estimation accuracy, especially, the combination of PLSR and continuum removal (R2=0.715, RMSE=0.218kg/m2). The experimental results demonstrated that PLSR is recommended for highly multi-collinear data sets. The combination of continuum removal and PLSR could improve the estimation accuracy of winter wheat biomass.