This quasi-experimental investigation describes the influence of text signals on second language expository science text comprehension. In two course sections, mixed proficiency Korean English language learners (n = 88) read one of two print-based English expository text passage versions. Participants in one section (n = 44) were given a version with interesting but "non-important subtopics" (NIS) underlined, while participants in the other section (n = 44) were given an alternate version with the same number of "substantively important subtopics" (SIS) underlined. Participants read the text passage and created a visual map of the text, and then completed a comprehension posttest that measured global inferences (all in English). Analysis of variance of the comprehension posttest data revealed significant differences for the two main factors, proficiency level (Low or High) and text signal condition (NIS or SIS), and for the interaction of proficiency and text signal condition; with Cohen effect sizes, d = Low-NIS (0.16) < High-NIS (0.56) < Low-SIS (0.74) < High-SIS (1.37). Descriptively, both headings and important subtopic terms predominated in the SIS maps, but only non-important subtopic terms and not headings predominated in the NIS maps. Further, the visual map forms, as measured by vector pattern matching and by graph centrality, were also substantially different. Compared to the NIS maps, on average the SIS map forms were more relational and more like the expert's map, while the NIS map forms were more linear, and showed a primacy effect. These results suggest that coherent text signals in these print-based readings strongly influenced bilinguals' science expository text comprehension.