It is significant to mine Top-K most-frequent patterns in dataset. The existing algorithms usually use the k-most frequent items as the initial items, and use the support of item with lowest frequency in initial items as the initial border support. In fact, since the number of items in Top-K most-frequent patterns is much less than k, the efficiency of the existing algorithms is restricted. To solve this problem, an efficient mixed-searching based algorithm for mining Top-K most-frequent patterns, MTKFP is presented. The algorithm firstly mines some short item sets by breadth-first searching, and uses short item sets to obtain the scope of the initial items (the number of initial items is less than k) and the higher initial border support; then it obtains all Top-K most-frequent patterns by depth-first searching. The experimental results show that the number of initial items of MTKFP is 70% lower than that of existing algorithms, and the initial border support of MTKFP is higher than that of existing algorithms. Hence the performance of MTKFP is superior to that of the best existing algorithm.