Skilled readers were trained to recognise either the oral (n=44) or visual form (n=40) of a set of 32 novel words (oral and visual instantiation, respectively). Training involved learning the "meanings" for the instantiated words and was followed by a visual lexical decision task in which the instantiated words were mixed with real English words and untrained pseudowords, and the instantiated words were to be considered as words. The phonology-to-orthography consistency (feedback consistency) of the instantiated words was manipulated to investigate the role of feedback from phonology in orthographic learning. Masked consonant and vowel-preserving form primes were used in the lexical decision task as probes of orthographic learning. Feedback-consistent instantiated words were recognised significantly faster in lexical decision than feedback-inconsistent instantiated words, and facilitation was significantly greater from consonant-preserving than vowel-preserving primes for orally but not visually instantiated words. The results support the hypothesis that orthographic representations based on a consonant frame can be generated from the speech signal before encountering the printed forms, and that feedback from phonology is involved in the early stages of orthographic learning.