A biometric algorithm for personal recognition by identification or verification of the person from his/her Arabic handwriting is presented. The system is online and text-independent. Dynamic data pertaining to handwriting, such as pressure, altitude and azimuth is collected from the user during the writing process, and is used to train models for each writer, after which a recognition system is developed for implementation. The focus in this work is to achieve a high recognition rate with minimal requirements for data and processing time. Based on our experimental results , a system based on Gaussian mixture models (GMM) yields the best classification performance. A full system is developed with a MATLAB-based Graphical User Interface and the Wacom Intuos4 digitizer tablet. We have achieved a best identification rate of 96% for writer identification; whereas for verification, we have obtained an FAR of 4.26% and an FRR of 4.00 %.