How many people are detained by ICE annually, and what percentage are later deported?
Executive summary
ICE’s footprint is large but fragmented in public data: the agency carries a point-in-time detained population measured in the tens of thousands and conducts roughly six-figured annual removals, but available datasets do not supply a single, authoritative numerator and denominator to produce one definitive “what percentage of detained people are later deported” figure [1] [2] [3]. Public sources show ICE averaged about 146,000 removals per year in FY2020–24 while the number of people detained at any one time has ranged from roughly 41,000 to more than 68,000 in 2025; how many distinct individuals pass through detention across a full year is larger and less consistently reported [3] [1] [2] [4].
1. How many people are held in ICE custody — snapshots versus flows
Point-in-time counts and flow counts tell two different stories: ICE’s regularly published “currently detained” reports put the detained population in 2025 anywhere from about 41,000 (early 2025) to more than 68,000 in December 2025, reflecting rapid increases in that year [1] [4] [2]. Those figures capture who is behind bars on a given day; separate ICE and third‑party datasets show monthly and quarterly “book‑ins” and releases that imply many more distinct people move through custody over a full year—for example, ICE reported 41,624 initial book‑ins in October 2025 alone, an indication that annual inflows are measured in the hundreds of thousands [2] [5].
2. How many people does ICE deport each year?
Research organizations and DHS reporting converge on a clear ballpark: DHS averaged roughly 352,000 deportations per year across FY2020–24, while Migration Policy’s analysis attributes an average of about 146,000 of those annual removals to ICE specifically over that period [3]. ICE’s own quarterly releases also document tens of thousands of removals in single quarters—for instance, ICE reported nearly 68,000 removals in the third quarter of FY2024, underscoring that removals under ICE’s authority occur at scale [6].
3. What percentage of detained people are later deported — why the data don’t yield a single answer
A single percentage requires linking the set of unique individuals detained during a year to the set later removed; public sources do not supply a clean, agency‑published crosswalk. ICE and DHS publish point‑in‑time detention counts, quarterly removals, and monthly book‑ins, but these datasets are produced on different schedules, with different definitions (CBP versus ICE custody, book‑ins versus arrests versus encounters), and are known to be revised and incomplete in places, which prevents a definitive published “of those detained, X% were deported” figure [7] [8] [5]. Independent archivists and researchers who have FOIA datasets (Deportation Data Project, Vera Institute) have assembled more granular data showing large flows through the system, but those reconstructions still stop short of producing a single, uncontested conversion rate because ICE does not routinely publish a consolidated annual matched cohort linking each detention episode to a final outcome [5] [9].
4. Reasonable, bounded inferences from the available numbers
Using only the numbers that the agencies and major analyses publish: ICE averages roughly 146,000 removals per year [3], while tens of thousands are detained at any given time and monthly book‑ins can exceed 40,000, implying annual detained‑person flows likely in the low‑to‑mid hundreds of thousands [2] [1]. Under those conditions, any back‑of‑the‑envelope conversion rate from “people booked into ICE custody during a year” to “ICE removals in that same year” would be highly sensitive to definitional choices and could plausibly range from well under 50 percent to higher, depending on whether one counts CBP‑handled repatriations, people released to alternatives to detention, administrative returns, or multi‑year cases [3] [2] [5]. Because sources disagree on scope and because ICE’s public tables are partial and periodically revised, producing a single authoritative percentage would be misleading [7] [9].
**