We ran both the multivariate linear regression and conditional quantile regression on the data collected from a 17-week flipped College English intervention so as to measure and decompose its effect on students’ L2 acquisition. Results show that the flipped intervention significantly improves learners’ language performance. Specifically, the learning returns in the flipped context is through reading, listening and writing in the rank of weakening importance, with the growth convergence surfacing at the high end of the spectrum, which indicates that the flipped classroom helps to build up learners’ basic language knowledge but fails to improve their deep-level language processing capacity. We also find that the flipped intervention also causes interval effects as its efficiency stems from the improvement of the performance of students at middle- and lower- level while high achievers show limited improvements.