Hyperspectral Image (HSI) contains a large number of spectral bands that easily results in the curse of dimensionality. The traditional classification methods just apply the spectral information while they ignore the spatial information. To address this problem, a dimensionality reduction algorithm combining Weighted Mean Filter (WMF) and Manifold Reconstruction Preserving Embedding (MRPE) was proposed in this paper. According to the spatial consistency property of HSI, firstly, the method applies WMF to all pixels which can reduce the spectral difference of data points from the same class. Then, the weights of the spatial neighbor points are enhanced in manifold reconstruction. This method effectively extracts the discriminant features and achieves the dimensionality reduction. Experimental results on PaviaU and Urban data sets show that the proposed method has better classification accuracy than other algorithms. When 5% and 1% of training samples were randomly selected from the two data sets, the overall accuracies based on MRPE can reach 98.76% and 80.21%. The proposed method enhances the low-dimensional manifold representation with the spatial information and improves the performance of HSI classification.