The aim of this thesis was to test whether the amount of self-corrections would influence the quality of the translated text. Study was conducted on nine students of translation studies who translated texts of intermediate difficulty in keylogging software Translog II from English into Croatian. After self-corrections were categorised and counted, it was found that the most frequent types of self-corrections were word substitution and spelling, which are thought to be the result of student’s over-editing and insecurity. Translations were evaluated using TAUS DQF model, after which evaluation scores were related to the number of self-corrections. Using Pearson test of correlation, it was found that there was no correlation with regards to the amount of self-corrections and evaluation score, which may be due to the small and diverse group of students who may differ greatly in translation experience. The result could also point to many students not being acquainted with normative rules of Croatian language. Time and number of self-corrections were found to be correlated, and evaluation score and amount of time were approaching significance, but were not correlated. This kind of translation process research may help in developing computer-assisted translating tools and help in understanding translator’s mind. It is suggested that future studies include a greater number of participants and test amount of self-corrections and translation quality in professional translators.