The perception and production of non-native speech sounds is the key to learning a new language. The differences between the native and the target language sound systems cause learning problems, but orthographic conventions may also affect the learning process. We tested whether a misleading orthography in contrast to phonemic transcription affects the manner in which native Finns learn to produce a non-native speech sound embedded in a pseudo word context. After the two day training protocol, the subjects who, in addition to the acoustic stimulation, were exposed to transcription cues, significantly changed their non-native productions according to the target. In contrast, the subjects who trained with the orthographic stimuli, changed their productions away from the acoustic target and towards the visual one. This result suggests that visual information is of crucial importance in learning to modulate articulation according to the target language model.