Simultaneous interpreting (SI) is a complex bilingual verbal activity that poses great challenges for working memory (WM) and language proficiency. Fluency is one of the crucial indicators in evaluating SI quality, the violation of which is characterized by disfluency indicators such as interruptions, hesitations, repetitions, corrections, and blanks. To uncover factors underlying fluency in SI, 22 interpreting students performed a battery of tasks to test their language proficiency and WM. Two SI tasks, both from Chinese to English and from English to Chinese, were also conducted, and fluency was evaluated according to the five indicators. Two factors (language proficiency and WM) and the five objectively measured disfluency indicators were then used as input for a regression analysis in both directions to model factors underlying fluency in SI performance. The results reveal that, with fluency measured as a whole, WM and directionality yield a significant effect on fluency, and that WM is the only variable that predicts fluency in both directions, accounting for 50 and 51% of the variation in the occurrence of disfluencies in Chinese–English and English–Chinese interpreting, respectively. The findings clarify for the first time the role of language proficiency, WM, and directionality upon fluency in SI, indicating the critical role of WM capability as compared with language skills in fluent production. The research also supports the position that, for interpreting students, interpreting performance tends to be more fluent in the non-native to native language direction.