A series of experiments explored the role of subvocalisation in fluent reading. Experiment I showed that when subjects were required to suppress articulation while reading, their ability to detect anomalous words or errors of word order in prose was markedly impaired although speed of reading was unaffected. Experiment II showed that this decrement was not a general effect due to performing a secondary task, since a concurrent tapping task did not impair detection accuracy. A third study explored the role of acoustic interference in reading by requiring subjects to detect errors in prose while attempting to ignore irrelevant speech, with or without articulatory suppression. Once again articulatory suppression led to a clear decrement in the subject's ability to detect errors, while unattended speech had no effect on performance. None of the manipulations influenced the speed with which the subjects performed the reading task. It is concluded that subvocalisation allows the creation of a supplementary articulatory code which is produced and utilised in parallel with other aspects of reading. Such a code seems particularly suitable for monitoring order information.