The paper focuses on the problem of detecting sentences containing causal relations in Polish legal texts. The identification of these relationships and their decomposition is a key factor in the effective analysis of legal texts and an important aspect in the extraction of parts of such relationships. This represents a contribution to the development of the field for languages other than English. The paper presents an analysis of the created dataset and based on it, classification was performed in nine different experiments using selected machine learning and deep learning algorithms (including several large BART-type models), taking into account the specifics of legal language. The experiments confirm the effectiveness of the proposed method, where the best model detected sentences containing both explicit and implicit causality with an accuracy of approximately 86%. These results lead to further questions and point to further directions for future development, especially in the field of reasoning from legal texts.