This paper presents an efficient VLSI architecture for transposing matris in high speed. In the case of transposing N*N matrix, N$^{2}$ numbers of transposition cells are configured as regular and spuare shaped structure, and pipeline structure for operating each transposition cell in paralle. Transposition cell consists of register and input data selector. The characteristic of this architecture is that the data to be transposed are divided into several bundles of bits, then processed serially. Using the serial transposition of divided input data, hardware complexity of transpositioncell can be reduced, and routing between adjacent transposition cells can be simple. the proposed architecture is designed and implemented with 0.5 .mu.m VLSI library. As a result, it shows stable operation in 200 MHz and less hardware complexity than conventional architectures.