Although extensive reading is regarded by many practitioners as a potentially very useful means of assisting L2 development, experimental enquiry into its effectiveness has so far produced little more than a collection of somewhat disparate findings. Nor has any attempt been made to categorically link any such research findings with second language acquisition theory. Consequently, we have no coherent, research-based theory of L2 extensive reading. Using data from a large-scale project implemented in Hong Kong secondary schools, the L2 English writing of students participating in an extensive reading scheme as part of the school curriculum was compared to that of non-participant students. Samples of timed narrative writing from 392 students in Secondaries 2 and 3 were rated holistically on a scale of 1 – 6 for overall quality, grammatical complexity, grammatical accuracy, vocabulary range, coherence, spelling and conventions of presentation. A subset of 150 compositions from two control and two experimental classes were further evaluated on a range of objective measures. Results from the two evaluation procedures were cross-referenced, and indicate that extensive reading in an L2 may benefit language development in quite specific ways. Findings are discussed within the context of current psycholinguistic and neurolinguistic theory and an explanation consistent with such theory is proposed. It is argued that, because it is likely to be subserved by a different memory system from that which subserved formal classroom instruction, extensive reading may enhance levels of automaticity, thus favouring the development of fluency, and, concomitantly, complexity and coherence. At low levels of L2 competence, extensive reading may also accelerate the acquisition of basic grammar through frequency effects.