diff --git a/src/uucore/src/lib/features/parser/parse_signed_num.rs b/src/uucore/src/lib/features/parser/parse_signed_num.rs
index 836c4f0a05..f0465c9d63 100644
--- a/src/uucore/src/lib/features/parser/parse_signed_num.rs
+++ b/src/uucore/src/lib/features/parser/parse_signed_num.rs
@@ -9,6 +9,7 @@
 //! sign indicates different behavior (e.g., "first N" vs "last N" vs "starting from N").
 
 use super::parse_size::{ParseSizeError, parse_size_u64, parse_size_u64_max, size_offset};
+use crate::display::Quotable;
 
 /// The sign prefix found on a numeric argument.
 #[derive(Debug, Clone, Copy, PartialEq, Eq)]
@@ -95,10 +96,10 @@ pub fn parse_signed_num_max(src: &str) -> Result<SignedNum, ParseSizeError> {
         // Otherwise "0K" would parse as 1KiB (bare suffix means 1).
         // A genuinely bare suffix with no digits at all (e.g. "kiB")
         // still parses as 1 of that unit.
-        parse_size_u64_max(trimmed)?;
+        parse_size_u64_max(trimmed).map_err(|e| as_typed(e, size_string))?;
         0
     } else {
-        parse_size_u64_max(trimmed)?
+        parse_size_u64_max(trimmed).map_err(|e| as_typed(e, size_string))?
     };
 
     Ok(SignedNum { value, sign })
@@ -141,6 +142,28 @@ pub fn number_offset(src: &str) -> usize {
     size_offset(src, |c| matches!(c, '+' | '-'))
 }
 
+/// Put back the leading zeros the parser stripped, so the error names the
+/// argument the way it was typed.
+///
+/// Zeros are only removed so the number is read as decimal rather than octal,
+/// which is an implementation detail the message should not leak: GNU reports
+/// `tail: invalid number of bytes: '007z'`, not `'7z'`. The sign is left off,
+/// also matching GNU, which reports `-c-0fb` as `'0fb'`.
+fn as_typed(error: ParseSizeError, size_string: &str) -> ParseSizeError {
+    let quoted = format!("{}", size_string.quote());
+    match error {
+        // These two carry the quoted operand and nothing else, so it can be
+        // swapped for the one that was actually typed.
+        ParseSizeError::InvalidSuffix(_) => ParseSizeError::InvalidSuffix(quoted),
+        ParseSizeError::ParseFailure(_) => ParseSizeError::ParseFailure(quoted),
+        // `SizeTooBig` carries an explanation after the operand and
+        // `PhysicalMem` is not about the operand at all, so neither can be
+        // rebuilt from the string alone. `parse_size_u64_max` clamps instead
+        // of overflowing, so neither reaches this in practice.
+        other => other,
+    }
+}
+
 /// Strip the sign prefix from a string and return both the sign and remaining string.
 fn strip_sign_prefix(src: &str) -> (Option<SignPrefix>, &str) {
     let trimmed = src.trim();
@@ -182,6 +205,31 @@ mod tests {
         assert_eq!(&operand[at..][1..3], "fb");
     }
 
+    /// GNU names the argument as it was typed. The leading zeros are stripped
+    /// only so the number is read as decimal rather than octal, and that
+    /// detail must not reach the message: GNU reports `'007z'`, not `'7z'`.
+    #[test]
+    fn an_invalid_count_is_reported_with_its_leading_zeros() {
+        for operand in ["0fb", "00x", "000ff", "0abc"] {
+            let error = parse_signed_num_max(operand).unwrap_err();
+            assert!(
+                error.to_string().contains(&format!("'{operand}'")),
+                "{operand} was reported as {error}"
+            );
+        }
+    }
+
+    /// The sign is not restored along with the zeros: GNU reports the operand
+    /// of `tail -c-0fb` as `'0fb'`.
+    #[test]
+    fn the_sign_is_left_off_the_reported_count() {
+        let error = parse_signed_num_max("-0fb").unwrap_err();
+        assert!(
+            error.to_string().contains("'0fb'"),
+            "-0fb was reported as {error}"
+        );
+    }
+
     #[test]
     fn test_no_sign() {
         let result = parse_signed_num_max("10").unwrap();
diff --git a/tests/by-util/test_head.rs b/tests/by-util/test_head.rs
index 7ce7dac6b1..3f9446b714 100644
--- a/tests/by-util/test_head.rs
+++ b/tests/by-util/test_head.rs
@@ -1181,3 +1181,17 @@ head: invalid number of bytes: '1fb'
             .stderr_is("head: invalid number of bytes: '1fb'\n");
     }
 }
+
+#[test]
+fn test_invalid_count_keeps_its_leading_zeros() {
+    // Leading zeros are stripped only so the count is read as decimal rather
+    // than octal. That is internal, so GNU still names the argument as typed.
+    new_ucmd!()
+        .args(&["-c", "0fb", "/dev/null"])
+        .fails_with_code(1)
+        .stderr_is("head: invalid number of bytes: '0fb'\n");
+    new_ucmd!()
+        .args(&["-n", "00x", "/dev/null"])
+        .fails_with_code(1)
+        .stderr_is("head: invalid number of lines: '00x'\n");
+}
diff --git a/tests/by-util/test_tail.rs b/tests/by-util/test_tail.rs
index a935b69db6..21bda5db06 100644
--- a/tests/by-util/test_tail.rs
+++ b/tests/by-util/test_tail.rs
@@ -5277,3 +5277,22 @@ mod diagnostics {
             .stderr_is("tail: invalid number of lines: '5QQ'\n");
     }
 }
+
+#[test]
+fn test_invalid_count_keeps_its_leading_zeros() {
+    // Leading zeros are stripped only so the count is read as decimal rather
+    // than octal. That is internal, so GNU still names the argument as typed.
+    new_ucmd!()
+        .args(&["-c", "0fb", "/dev/null"])
+        .fails_with_code(1)
+        .stderr_is("tail: invalid number of bytes: '0fb'\n");
+    new_ucmd!()
+        .args(&["-n", "000ff", "/dev/null"])
+        .fails_with_code(1)
+        .stderr_is("tail: invalid number of lines: '000ff'\n");
+    // The sign is not put back with them: GNU reports `-c-0fb` as '0fb'.
+    new_ucmd!()
+        .args(&["-c-0fb", "/dev/null"])
+        .fails_with_code(1)
+        .stderr_is("tail: invalid number of bytes: '0fb'\n");
+}