Abstract: A fast set partitioning in hierarchical tree (SPIHT) algorithm is proposed in this paper. Compared to the typical SPIHT algorithm, three measures have been taken to speed up SPIHT algorithm. First, human visual system (HVS) is applied to weight the wavelet coefficients, and thus the impacts of subjective fators on image coding are paid more attention. Second, a list of maximum pixels is calculated and introduced so as to, in the process of scanning coefficients, only the corresponding data in such list is necessary to be compared with the threshold. Third, more than one spatial orientation trees are parallel encoded. Experimental results show that the modified SPIHT has double implementation efficiency in terms of execution time, while the reconstructed image has better visual quality with an average gain of 0.2dB PSNR.