Ability grouping, organizing classes homogeneously by L2 proficiency, has been commonly used in Taiwanese English as a Foreign Language (EFL) classes. This quasi-experimental (within-subjects design) study examined proficiency gains of 785 Taiwanese university students over three years enrolled in a general English (GE) program that employed ability grouping. The standardized test used for this study was the "College Students English Proficiency Test" (CSEPT). The results indicated students gained in English proficiency over time, from entry into the program to their last year of English instruction. Further post hoc analysis of the long-term proficiency changes showed that students with an observed A2 (CEFR) proficiency, upon entry, had more pronounced gains, over the three years, than their A1 and B1 counterparts. The study concluded that a leveled English curriculum maximized the learning experience for A2 level students and allowed them continuous proficiency gains. However, the fact that B1 level students did not show consistent progress is perhaps due to plateau effect when their test scores hit the graduation benchmark. As for the A1 students, their lack of achievement may be due to their low self-esteem. The pedagogical implication suggests the need to revisit the leveled (ability grouping) English curriculum for A1 and B1 level learners.