According to the basic theories of Brownian motion, a mixed search model is proposed. By dividing every search process into two phases: single random walk search and multiple random walk searches, it improves the efficiency of a single random walk while reduces the hardware cost of multiple random walks. Compared with the mixing navigation model proposed by Tao Zhou, our model converges much faster on complex networks with lower hardware cost. We also compared two selection strategies: random selection and target selection, and found that our model improves the search efficiency further on complex networks when employed with target selection, which is prefered to the nodes of the highest degrees. Besides, by simulations on scale-free networks, we found the efficiency of our model is far higher than the single random walk and comparable to multiple random walks while the hardware cost of our model is far less. Finally, we give an absorption strategy to reduce the traffic load produced by the MS model.