Vector quantization (VQ) is a popular signal compression method. In a VQ framework, the encoding speed is one of the key issues for its practical applications. In principle, the high dimension of the original (n×n)-dimensional vectors is the reason that makes it is computationally very expensive to find the best-matched codeword for a given input vector by computing Euclidean distances. As a result, a lot of fast VQ encoding methods have been developed in the previous works based on using scalar statistical features (i.e., sum, variance or L2 norm) or lower dimensional multi-resolution representation (i.e., various pyramid data structures) of a vector to deal with this computational complexity problem caused by the high dimension of the original vectors. In particular, a very effective mixed-pyramid data structure is reported in the previous work , which features a minimum configuration by combining a 2-PM core sum pyramid and a (n×n)-PM auxiliary variance pyramid together. However, this mixed-pyramid failed to take the element order of a vector into account. In order to further improve the encoding performance of the previous work , this paper proposes to introduce a dynamic element reordering operation to let all elements of a codeword rearranged in an ascending order by an offline sorting process before it goes to construct the mixed-pyramid for the codeword. After the element reordering, it is possible to make the intermediate levels in the mixed-pyramid of a codeword keep more energy and become more powerful for rejection tests. Experimental results confirmed that the encoding efficiency of the proposed method remarkably outperforms the previous work .