Upon examination of the history of development of machine translation, it becomes clear that the ever-growing demand for rapid translation has been surpassing the human translators' capacity. Furthermore, the expansion of the Internet has created a demand for prompt online translations, which human translators cannot carry out. Statistically based tool Google Translate provides a multilingual machine translation by automatically translating one written language to another. The purpose of this research is to evaluate Google Translate machine translation output for English – Croatian and Croatian – English language pairs in four genres. The corpora sample is a collection of eight text extracts from four different text genres. Machine translation outputs were analysed by human evaluation on the criteria of fluency and adequacy. Human evaluation is enhanced by detailed error analysis in order to examine the error types for their classification according to text genre. The findings may be useful in further research. The results show that Google Translate is still limited in translation of certain text types, especially those with complex structures and grammatical concepts, rich morphological and semantic system. In addition, the most notable error is mistranslation of multi-word expressions, but there is also a high frequency of both inflection and preposition errors.