In GPS software receiver, there are massive correlating operations which cost huge time and hardware resources in the acquisition. The GPS software receiver uses the FFT algorithm to decrease process time obviously. However, this method is used limitedly on the embedded devices due to the limited hardware resources and time-consuming. Thus, this paper proposes an adaptive search algorithm. This algorithm makes a partition of different levels of steps and uses different steps to improve the efficiency of the acquisition. This method implements multi-level steps search in finding the 2-D spectrum peak of circular cross-correlations. In the search process, it preferres to big steps to reduce search operations as far as possible. Because the search process of every satellite uses multi-level steps other than the steps of conventional partition which is lowest level in the multi-level steps, it can remove FFT operations obviously. And to avoid unnecessary computation, there are some tables to restore the various levels of steps. This method reduces redundancy FFT operations in the acquisition and improves the efficiency of the acquisition. Experimental results show this method has improved the efficiency of the acquisition and decreased the process time significantly.