We examined effects of error production in training of spelling in four individuals with stroke-induced dysgraphia. In a single participant crossover design, we provided errorless and errorful spelling training in counterbalanced phases. Improvements in spelling to dictation (large effect sizes) were evident for trained words in all participants following both errorful and errorless training phases, with some advantage of errorful over errorless for three of four participants. Little generalisation was noted for untrained control words. Post-treatment improvements were also noted in a standardised writing test in three of four participants. At one month after training completion, gains were largely maintained for trained words, again with larger effect sizes for errorful than errorless words in three of four participants. However, despite benefits for errorful training, participants stated a preference for the less frustrating errorless training protocol.