It is well known that people's movement exhibits some pattern. Predicting people's location opens possibilities for novel applications. For example, a local-community crowdsourcing app could offer people tasks to help other while they are in an appropriate location or on route to such. Traditional location prediction methods mainly focus on important places prediction without considering the routes user takes. In this talk, we will present the design and implementation of a two-phase location prediction approach, SmartLoc, which considers not only the important places, but also the routes. Further, we have implemented a spatiotemporal context-based algorithm to extract the important places and a route-matching algorithm to extract the real routes, respectively. We evaluate our approach with a real-world dataset involving 21 users in a period of three to 70 days. The results demonstrate that the prediction accuracy is higher than 90% for people who have two or three important places and routes. For people who have more than 10 important places and 20 routes, our approach can even achieve the prediction with accuracy not lower than 70%.