This study explored the nonlinear pattern by which language knowledge moderates the effect of background knowledge on Language for Specific Purposes (LSP) reading performance and the number of language thresholds that locate the turns of such an effect on background knowledge. We tested two hypothesized patterns in which students’ language knowledge interferes with background knowledge: a linear pattern and a quadratic pattern (background knowledge effect moves up-then-down with the increase of language knowledge). A total of 1,491 nursing students from eight medical colleges in China participated in the study. Their background knowledge, language knowledge and LSP reading ability were measured using a nursing knowledge test, an English grammar knowledge test, and a nursing English reading test, respectively. Students’ response data were first scored using multidimensional item response theory and then modeled using the innovative method of multi-layered moderation analysis (MLMA). The results supported the quadratic moderation pattern, which we labeled as island ridge curve (IRC). On the IRC, two language thresholds emerged: a resurfacing threshold (theta = -1.49; from where the positive effect of background knowledge emerged and started to increase) and a downhill threshold (theta= 1.00; from where the maximum effect of background knowledge emerged and started to decrease).