In this paper we present a wearable wireless accelerometer device for people fall detection in ambient assisted living (AAL) applications, communicating with care holders and relatives of the assisted person through an ADSL based gateway. The wearable system is a PCB with a commercial three-axes, digital-output MEMS accelerometer chip, a ZigBee transceiver and a programmable logic (FPGA). The raw acceleration stream from the MEMS sensor is processed by suitable algorithms on the FPGA, whose aim is to detect potential falls and their level of confidence, delivering an alarm information to the gateway by means of the ZigBee transceiver. Several sessions of data collection with real actors have been performed to define and characterize fall detection algorithms. Experimental results show that the accelerometer is a fundamental component of a multi-sensor network system and on this basis a custom device has been realized.