Previous studies have investigated how second language (L2) readers construct memory for narrative texts according to causal relations between the events described. This study aims to extend their findings by including semantic text relations (similarity of meaning) as another variable, which are theoretically expected to play an additional role in comprehension. With this aim, 121 Japanese learners of English read a set of 200-word long narratives adapted from previous studies. The causal relationships between the text statements were identified using a causal network analysis, whereas the semantic text relations were evaluated by a computational method (latent semantic analysis). After reading the narratives, participants performed a written recall task. The results of the recall analysis confirmed that the global causal text relations influence the memory for narrative texts. The results further revealed that local semantic relations between adjacent text statements also have an impact. These findings are discussed in light of theories of discourse comprehension.