diff --git a/src/uu/sort/src/external_sort/mod.rs b/src/uu/sort/src/external_sort/mod.rs
index 725b17bbd..662250e1d 100644
--- a/src/uu/sort/src/external_sort/mod.rs
+++ b/src/uu/sort/src/external_sort/mod.rs
@@ -113,7 +113,7 @@ pub fn ext_sort(
 
         chunk.push(seq);
 
-        if total_read >= settings.buffer_size {
+        if total_read + chunk.len() * std::mem::size_of::<Line>() >= settings.buffer_size {
             super::sort_by(&mut chunk, &settings);
             write_chunk(
                 settings,
@@ -136,6 +136,9 @@ pub fn ext_sort(
         iter.chunks += 1;
     }
 
+    // We manually drop here to not go over our memory limit when we allocate below.
+    drop(chunk);
+
     // initialize buffers for each chunk
     //
     // Having a right sized buffer for each chunk for smallish values seems silly to me?