Abstract The purpose of this study is to look at possible strategies of evaluating machine translations and to review the capabilities of current MT systems. It also aims to measure the level of equivalence that could be supported by the machine translation systems and to compare the translation quality of different types of machine translation systems. After reviewing the background of MTs and the main-flow evaluation metrics, five latest machine translation systems (Systran, Google, PT2008, LEC Trans DotNet, and Kingsoft AIT) have been tested by assessing their translations of five abstracts adopted from the Financial Analysts Journal. By analyzing the translation quality from the perspective of equivalence proposed by Mona Baker (1992), this research has found that machine translation can only partially achieve lower levels of equivalence for preset domains, such as equivalence at word level and equivalence at above word level. Although most systems claim to be full-text capable, they could barely deal with equivalence at contextual level and could only make appropriate translations to limited extent. Based on the performances of different systems, each layer of translation equivalence is discussed further. Several procedures in pre-editing are recommended for further exploration as they may possibly reduce the ambiguity of the original text.