Emerging ubiquitous and pervasive computing applications often need to know where things are physically located. To meet this need, many location-sensing systems have been developed, but none of the systems for the indoor environment have been widely adopted. In this paper we propose Proximity Mining, a new approach to build location information by mining sensor data. The Proximity Mining does not use geometric views for location modeling, but automatically discovers symbolic views by mining time series data from sensors which are placed in surroundings. We deal with trend curves representing time series sensor data, and use their topological characteristics to classify locations where the sensors are placed.