An important aspect of the creation of scientific text is metadiscourse — a combination of linguistic and rhetorical tools. The use of metadiscourse affects the relationship between the text’s author and reader, including attitudes about the text itself, creating a social act in which the text producer and recipient interact. This paper outlines that relationship. The goals of the paper are 1) to show the use of metadiscourse in Latvian scientific text research and 2) to show a selection of problems related to metadiscourse marker identification. Thirty scientific articles about Latvian linguistics were selected for this research. The material has been analysed using Ken Hyland’s metadiscourse analysis model. A part of the data underwent double coding. Hyland’s model is divided in two categories: interactive and interactional markers. Interactive markers help guide the reader through the text, whereas interactional markers involve the reader in the content of the text and arguments, attempting to convince the reader to agree with the author’s opinion. This research shows that the usage of metadiscourse markers is often unpredictable. The most commonly used markers are code glosses, which are used to add examples to the text. More attention should be given to the interactional markers, endophoric markers and frame markers, as these help the reader understand the author’s ideas. The text author’s opinion is shown via attitude markers, boosters, and hedges. The use of transition markers is pre-determined by the text type and evidentials are used when there are references to other work.substantial level of agreement between coders. However, the results also show that there are some problems in determining metadiscourse markers. Qualitative methods are best suited for metadiscourse analysis, while quantitative methods can be used to verify and compare data. After the double coding verification, it was found that the data has scored a 0.782 Cohen’s kappa, which shows a substantial level of agreement between coders. However, the results also show that there are some problems in determining metadiscourse markers. Qualitative methods are best suited for metadiscourse analysis, while quantitative methods can be used to verify and compare data.