Arabic words are constructed differently from European languages, and all words are based on a root that is formed from three consonants. There is a comparatively small number of these root forms, compared to English for example, but there is a much more elaborate and productive process of rule-based derivation to generate a huge variety of words. It has been claimed, although without real evidence to support this, that reading in L1 Arabic is a process of affix-stripping so that the tri-consonantal root can be recognised (Randall & Meara, 1991), and this provides an efficient route to word recognition and semantic meaning. This article reviews our recent research using an eye-tracker suggesting that this is occurring, and this process provides a route to word recognition which is even faster than that for L1 English speakers. As a process for reading English, or any other alphabetic system, this process is useless since there is no tri-consonantal root to recognise. This must impact on L2 learning and delay its development. It may also explain why these learners have a different structure to their L2 lexicon, compared with other learners, their comparatively poor reading comprehension performance, and their reliance on phonemic decoding.