In this paper, a two-tier classification system for online-recognition of handwritten Arabic letters is proposed based on dynamic time warping (DTW). The main feature set used here is the sequence of the vector angles obtained from the horizontal and vertical differential information of the handwriting sequence. We propose a general method of classification in both tiers to be minimum distance classifier, for which we explore some options for distance metrics including one based on DTW. Data collected from 30 different writers were used to validate the system's performance. Experimental results showed superior recognition rates with the proposed DTW-based distance metric. These rates are 94% at the superclass level and 89% at the sub-class level.