The impact of high (e.g. syntax and semantics) and low (graphemes) linguistic levels in 50 beginners' oral reading of running texts is explored, by means of a qualitative analysis of reading errors collected in a longitudinal study. The hypothesis, based on previous quantitative results from the same data, was that the graphemic and/or lexical levels could explain some of the reading errors, but that, for example, syntactic complexity or unexptected contexts could explain others. Unfamiliar words and words with compex graphemic structures did result in many reading errors. but a relatively large number of errors seem to be caused by, for example, syntactical or contextual constraints. In the longitudinal perspective, the readers tended to advance through various phases regarding their utilisation of higher linguistic levels, revealing a possible transition from a concentration mainly on the graphemic level via a dependency on and finally a sensitivity to context.