Recommending points of interest (POIs) to a user according to the user’s current location and past check-in activities is the focus in this paper. Previously proposed probabilistic and topic model-based methods predict the POIs based on the distribution of the POIs visited in the past, assuming that the next POI for the user follows the same distribution. Such methods tend to recommend the POIs in the cities or regions that the user has visited before because only such cities or regions have observed ratings for the user. Thus, these works are not suitable for a user who travels to a new city or region where she has not checked-in any POI previously. To address this issue, we distinguish the user preferences on the content of POIs from the user preferences on the POIs themselves. The former is long term and is independent of where POIs are located, and the latter is short term and is constrained by the proximity of the location of the POI and the user’s current location. This distinction motivates a location-independent modeling of user’s content preferences of POIs, and a location-aware modeling of user’s location preferences of POIs. The final recommendation of POIs is derived by combining the predicted rating on content and the predicted rating on location of POI. We evaluate this method using the Yelp and Foursquare data sets. This approach has superiority over the state-of-the-art and works well in the “new city” situation in which the user has not rated any of the POIs in the current region.