In order to quickly and non-destructively detect and assess mildew and moisture content of timber, the hyperspectral data of timber was actively acquired by hyperspectral light detection and ranging (LiDAR), and a method was designed to analyze mildew characteristics and establish timber moisture content prediction model. Firstly, timber sample(white pine) hyperspectral data was measured at monthly intervals for four months, and the spectral characteristics of mildew occurrence and development(normal, wet and mildew state) were analyzed. Then based on analyzing the spectral characteristics of sample moisture content, competitive adaptive reweighted sampling, successive projections algorithm and competitive adaptive reweighted sampling-successive projections algorithm were employed to extract feature wavelength. Finally, prediction models were established with partial least squares regression respectively. The results show that, the spectral reflectance of normal state is highest and the mildew’s is lowest; when the mildew state is stable, the spectral reflectance changes slowly with time and tends to stabilize; and the model based on combined algorithm achieved the best predictive performance, the correlation coefficient and root mean square error of the prediction set are 0.9073 and 0.7564, respectively. The active acquisition of hyperspectral information by hyerspectral LiDAR can be used to assess mildew and predict timber moisture content, providing new ideas for rapid non-destructive detecting of wood buildings.