I elicited speech errors by experiments and analysed them to see what they reveal about vowel harmony (VH). The basic question is whether VH helps Turkish speakers to learn an artificial language system. The question of whether the subjects learned VH was tested in three ways: first, the diversity and quantity of their speech errors; second, by the test phase and third by analyzing the words subjects remembered. The results revealed that subjects preserved the system they were trained on even in their speech errors. There were certain patterns in the speech errors of both harmony and disharmony subjects. This underlies the finding that as long as there is a pattern (whether harmony or disharmony) in the artificial language system, subjects were able to learn it. The backness/rounding harmony subjects were more successful than other groups. It was observed that height harmony was more difficult for Turkish subjects compared to backness/rounding harmony. But the results of the mixed condition subjects reveal that even vowel disharmony was easier than a lack of pattern. The results point out to the fact that VH is a property that facilitates both production and perception.