A personalizable fall detector system is presented in this paper. It relies on a semisupervised novelty detection technique and has been implemented in a smartphone application. Thus, it has been tested that the algorithm can run comfortably in this kind of devices. Details about the internal structure of the application and a preliminary evaluation are also shown. The main difference with previous approaches relies in the fact that semisupervised techniques only require activities of daily life for its operation. Departures from normal movements are considered as falls. In this way, no simulated falls are needed, except for testing the performance. Therefore, the system can be easily adapted to each user.