In this work, we train a text repair model as a post-processor for Neural Machine Translation (NMT). The goal of the repair model is to correct typical errors introduced by the translation process, and convert the "translationese" output into natural text. The repair model is trained on monolingual data that has been round-trip translated through English, to mimic errors that are similar to the ones introduced by NMT. Having a trained repair model, we apply it to the output of existing NMT systems. We run experiments for both the WMT18 English to German and the WMT16 English to Romanian task. Furthermore, we apply the repair model on the output of the top submissions of the most recent WMT evaluation campaigns. We see quality improvements on all tasks of up to 2.5 BLEU points.