The current study looks at whether bilinguals varying in language dominance show a processing advantage for idiomatic over non-idiomatic phrases and to what extent this effect is modulated by idiom transparency (i.e., the degree to which the idiom's figurative meaning can be inferred from its literal analysis) and cross-language similarity (i.e., the extent to which an idiom has an identical translation equivalent in another language). An eye tracking experiment was conducted in which Spanish-English bilinguals were presented with literally plausible (i.e., idioms that can be interpreted both figuratively and literally) transparent (e.g., "break the ice," where the figurative meaning can be deduced from analyzing the idiom literally) and opaque idioms (e.g., "hit the sack," where the meaning cannot be inferred from idiom constituents). Idioms varied along the dimension of cross-language similarity, with half the idioms having word for word translation equivalents in English and Spanish and another half being different, that is, having no similar counterpart in another language. Each idiom was used either in its literal (e.g., "get cold feet": become cold) or figurative meaning (e.g., "get cold feet": become afraid). In control phrases the last word of the idiom was replaced by a carefully matched control (e.g., "get cold hands"). Reading measures (fixation count, first pass/gaze reading time and total reading time) revealed that cross-language similarity interacts in an important way with idiom transparency, such that opaque idioms were more difficult to process than transparent ones, and different transparent idioms took faster to process than similar transparent idioms. Results are discussed with regard to the holistic vs. compositional views of idiom storage and the role of activated L1 (first language) knowledge in the course of L2 (second language) figurative processing.