Whereas most existing research has examined the prevalence of current English learners (ELs) in special education, we propose and test the use of the ever-EL framework, which holds the subgroup of EL students stable by following all students who enter school classified as ELs. Drawing on two administrative data sets, discrete-time hazard analyses show that whereas current EL students are overrepresented in special education at the secondary level, students who enter school as ELs are significantly underrepresented in special education overall and within most disability categories. Reclassification patterns, in part, explain these findings: EL students with disabilities are far less likely than those without disabilities to exit EL services, resulting in large proportions of dually identified students at the secondary level. These findings shed new light on EL under- and overrepresentation in special education and offer insights into policies and practices that can decrease EL special education disproportionality.