This study presents the first multifactorial corpus-based analysis of verb-particle constructions in a data sample comprising spoken and written productions by intermediate-level learners of English as a second language from 17 language backgrounds. We annotated 4,911 attestations retrieved from native speaker and language learner corpora for 14 predictors, including syntactic complexity, rhythmic and segment alternation, and the verb framing of the speaker's native language. A multifactorial prediction and deviation analysis using regression (Gries & Deshors, 2014), which stacks multiple regression analyses to compare native speaker and learner productions in identical contexts, revealed a complex picture in which processing demands, input effects, and native language typology jointly shape the degree to which learners' choices of constructions are nativelike or not.