This work investigates the feasibility of a personal verification system using gestures as biometric signatures. Gestures are captured by low-power, low-cost tri-axial accelerometers integrated into an expansion pack for palmtop computers. The objective of our study is to understand whether the mobile system can recognize its owner by how she/he performs a particular gesture, acting as a gesture signature. The signature can be used for obtaining access to the mobile device, but the handheld device can also act as an intelligent key to provide access to services in an ambient intelligence scenario. Sample gestures are analyzed and classified using supervised and unsupervised dimensionality reduction techniques. Results on a set of benchmark gestures performed by several individuals are encouraging.