Abstract: The fast algorithm of discrete wavelet transform (DWT) ensures that the wavelet transform is widely applied in signal processing. In the algorithm, the computation of values on sample space is very important. This paper, based on the discussion of several fast DWT algorithms, is focused in the computation methods of the values on sample space for prefilter selection, includingthewaveletmethod, the directmethod and the sampling function method; at the same time, the corresponding properties of each method are discussed. And then based on wavelet coefficients with minimal norm errors, an optimal design for the prefilter is put forward. The errors of the various methods mentioned above are compared by simulation; the results showthat the proposed optimal method has a fairly good precision in decomposition.