GPS map matching is the process to align observed GPS positions with road networks of a digital map. One of the key technique in a state-of-art GPS map matching algorithm is to select candidates for each observed GPS point. Traditional candidate selection algorithms focus on spatial proximity, which is not sufficient in real cases. This paper proposes a novel candidate selection algorithm for GPS map matching, called Space Map Matching (SMM). The SMM constructs a mapping relationship between space and road links based on GPS shifting patterns and driver preferences. Therefore, candidate selection is transformed from a spatial searching process into a mapping relationship looking-up process. Experiments on real datasets prove that the candidate selection algorithm proposed in this paper can outperform traditional algorithms in both accuracy and efficiency.