Abstract: As a vital component of human-computer interaction, gesture recognition has gained a lot of attentions in civilian applications recently. Many applications would benefit from such gesture recognition, e.g. smart phone, smart home system, Kinect, etc. In this work, we introduce an RFID (radio frequency identification) based solution to recognize gestures using COTS (commercial off-the-shelf) RFID tags and readers. The basic idea is treating signal features caused by perturbation of gestures as the fingerprint. Furthermore, we leverage multipath to increase the difficulty of matching, which provides a high recognition rate. Unlike past work, which required user to attach an RFID tag, our method does not require any device to be attached to users. Specifically, we solve the problem that RFID communication is discontinuous in time domain by data slicing. We then capture and extract characteristic matrix corresponding to each gesture by using a synthetic aperture radar (SAR). Finally, we adopt dynamic time warping (DTW) techniques to recognize gestures by matching with the priori gesture fingerprint database. We implement our method using a COTS RFID device and evaluate it with 6 commercial tags. Experimental results demonstrate that this device free method is feasible and it can achieve a correct recognition rate of about 85%.