The task of discovering prevalent co-location patterns (PCPs) from spatial data sets, i.e., discovering groups of spatial features with their instances frequently occurring together in the neighborhood of each other is an important branch of data mining. Although many algorithms for mining PCPs have been proposed, each of them has different advantages and disadvantages. Among them, the clique-based mining algorithms can give good performance in discovering long-size PCPs. While the level-wise search-based algorithms show better performance in generating short-size PCPs. However, the number of middle-size patterns often accounts for a larger proportion, therefore, the above two mining methods are hard to deal with. This work proposed an efficient clique-based level-wise search algorithm for discovering PCPs that combines the advantages of the two above mining methods. First, similar to the clique-based approach, the neighbor relationships of all instances are enumerated and arranged into a compact hash table structure. Secondly, the level-wise search method is used to search for PCPs on the hash table structure from size-2 candidates. The mining process is completed when no higher-size candidates are generated. The proposed algorithm is evaluated experimentally on a large number of synthetic and real spatial data sets. The experimental results show that this algorithm can give better performance than the state-of-the-art algorithms.