diff --git a/bindings/python/tests/bindings/test_normalizers.py b/bindings/python/tests/bindings/test_normalizers.py
index 99ab07d393..f2500eac39 100644
--- a/bindings/python/tests/bindings/test_normalizers.py
+++ b/bindings/python/tests/bindings/test_normalizers.py
@@ -2,7 +2,7 @@
 
 import pytest
 
-from tokenizers import NormalizedString
+from tokenizers import NormalizedString, Regex
 from tokenizers.normalizers import (
     BertNormalizer,
     Lowercase,
@@ -201,6 +201,23 @@ def test_can_modify(self):
         assert normalizer.prepend == "-"
 
 
+class TestReplace:
+    def test_replace_with_groups(self):
+        normalizer = Replace(Regex(r"(l)(e)"), r"$1 $2")
+        result = normalizer.normalize_str("le travail")
+        assert result == "l e travail"
+
+    def test_replace_with_group_zero(self):
+        normalizer = Replace(Regex(r"(\w+)"), r"[$0]")
+        result = normalizer.normalize_str("hello world")
+        assert result == "[hello] [world]"
+
+    def test_replace_no_capture_unchanged(self):
+        normalizer = Replace(Regex(r"\s+"), " ")
+        result = normalizer.normalize_str("hello   world")
+        assert result == "hello world"
+
+
 class TestCustomNormalizer:
     class BadCustomNormalizer:
         def normalize(self, normalized, wrong):
diff --git a/tokenizers/src/normalizers/replace.rs b/tokenizers/src/normalizers/replace.rs
index 5657574830..9bf48471b6 100644
--- a/tokenizers/src/normalizers/replace.rs
+++ b/tokenizers/src/normalizers/replace.rs
@@ -49,6 +49,8 @@ pub struct Replace {
     pub content: String,
     #[serde(skip)]
     regex: SysRegex,
+    #[serde(skip)]
+    expansion_re: Option<regex::Regex>,
 }
 
 impl Clone for Replace {
@@ -70,38 +72,48 @@ impl Replace {
             ReplacePattern::String(s) => SysRegex::new(&regex::escape(s))?,
             ReplacePattern::Regex(r) => SysRegex::new(r)?,
         };
+        let expansion_re = match &pattern {
+            ReplacePattern::String(_) => None,
+            ReplacePattern::Regex(r) => regex::Regex::new(r).ok(),
+        };
 
         Ok(Self {
             pattern,
             content: content.into(),
             regex,
+            expansion_re,
         })
     }
 }
 
 impl Normalizer for Replace {
     fn normalize(&self, normalized: &mut NormalizedString) -> Result<()> {
-        normalized.replace(&self.regex, &self.content)
+        normalized.replace_regex(&self.regex, self.expansion_re.as_ref(), &self.content)
     }
 }
 
 impl Decoder for Replace {
     fn decode_chain(&self, tokens: Vec<String>) -> Result<Vec<String>> {
-        tokens
-            .into_iter()
-            .map(|token| -> Result<String> {
-                let mut new_token = "".to_string();
-
-                for ((start, stop), is_match) in (&self.regex).find_matches(&token)? {
-                    if is_match {
-                        new_token.push_str(&self.content);
-                    } else {
-                        new_token.push_str(&token[start..stop]);
+        match &self.expansion_re {
+            Some(re) => tokens
+                .into_iter()
+                .map(|token| Ok(re.replace_all(&token, &self.content).to_string()))
+                .collect(),
+            None => tokens
+                .into_iter()
+                .map(|token| {
+                    let mut new_token = String::new();
+                    for ((start, stop), is_match) in (&self.regex).find_matches(&token)? {
+                        if is_match {
+                            new_token.push_str(&self.content);
+                        } else {
+                            new_token.push_str(&token[start..stop]);
+                        }
                     }
-                }
-                Ok(new_token)
-            })
-            .collect()
+                    Ok(new_token)
+                })
+                .collect(),
+        }
     }
 }
 
@@ -156,4 +168,28 @@ mod tests {
             vec!["hello", " hello"]
         );
     }
+
+    #[test]
+    fn test_replace_regex_groups_dollar() {
+        let original = "le travail";
+        let expected = "l e travail";
+
+        let mut n = NormalizedString::from(original);
+        Replace::new(ReplacePattern::Regex(r"(l)(e)".into()), r"$1 $2")
+            .unwrap()
+            .normalize(&mut n)
+            .unwrap();
+
+        assert_eq!(&n.get(), &expected);
+    }
+
+    #[test]
+    fn test_replace_regex_groups_decode() {
+        let tokens = vec!["le".to_string(), "test".to_string()];
+        let replace = Replace::new(ReplacePattern::Regex(r"(l)(e)".into()), r"$1 $2").unwrap();
+        assert_eq!(
+            replace.decode_chain(tokens).unwrap(),
+            vec!["l e", "test"]
+        );
+    }
 }
diff --git a/tokenizers/src/tokenizer/normalizer.rs b/tokenizers/src/tokenizer/normalizer.rs
index 5bebd5f7b4..61b825f6fd 100644
--- a/tokenizers/src/tokenizer/normalizer.rs
+++ b/tokenizers/src/tokenizer/normalizer.rs
@@ -1,4 +1,5 @@
 use crate::pattern::Pattern;
+use crate::utils::SysRegex;
 use crate::{Offsets, Result};
 use std::ops::{Bound, RangeBounds};
 use unicode_normalization_alignments::UnicodeNormalization;
@@ -673,6 +674,106 @@ impl NormalizedString {
         Ok(())
     }
 
+    /// Replace anything that matches the given regex with the given content,
+    /// supporting backreferences (`$1`, `$2`, etc.) in the replacement string.
+    pub fn replace_regex(
+        &mut self,
+        regex: &SysRegex,
+        expansion_re: Option<&regex::Regex>,
+        content: &str,
+    ) -> Result<()> {
+        let mut new_normalized = String::with_capacity(self.normalized.len()); // Initially allocate for the input size
+        let mut new_alignments: Vec<(usize, usize)> = Vec::with_capacity(self.alignments.len());
+        let mut last_end = 0; // Keep track of the last end position
+
+        for (start, end) in regex.find_iter(&self.normalized) {
+            let range = start..end;
+
+            let matched = &self.normalized[start..end];
+            let expanded = match (matched.is_empty(), expansion_re) {
+                (true, _) | (_, None) => content.to_string(),
+                (false, Some(re)) => re.replacen(matched, 1, content).to_string(),
+            };
+
+            let removed_chars = self.normalized[range.clone()].chars().count();
+
+            // Copy the part of the string that is before the match
+            new_normalized.push_str(&self.normalized[last_end..start]);
+            new_alignments.extend(self.alignments[last_end..start].iter().cloned());
+
+            let n_range = Range::Normalized(range).into_full_range(self.len());
+
+            // Retrieve the original characters that are being replaced. This let us
+            // compute the change in byte sizes along the way.
+            let mut replaced_normalized = self.normalized[n_range.clone()]
+                .chars()
+                .collect::<Vec<_>>()
+                .into_iter();
+            let initial_removed: usize = (&mut replaced_normalized)
+                .take(removed_chars)
+                .map(|c| c.len_utf8())
+                .sum();
+
+            let mut offset = (initial_removed + n_range.start) as isize;
+            let normalized = expanded
+                .chars()
+                .map(|c| (c, 1))
+                .map(|(c, changes): (char, i32)| {
+                    let idx = offset as usize;
+                    let align = if changes.is_positive() {
+                        if idx < 1 {
+                            (0, 0)
+                        } else {
+                            // This is a newly inserted character, so it shares the same alignment
+                            // than the previous one
+                            self.alignments[idx - 1]
+                        }
+                    } else {
+                        self.alignments[idx]
+                    };
+
+                    // If we are replacing a character, find it and compute the change in size
+                    let replaced_char = if !changes.is_positive() {
+                        replaced_normalized.next()
+                    } else {
+                        None
+                    };
+                    let replaced_char_size = replaced_char.map_or(0, |c| c.len_utf8());
+
+                    // If we are removing some characters, find them too
+                    let total_bytes_to_remove = if changes.is_negative() {
+                        (&mut replaced_normalized)
+                            .take(-changes as usize)
+                            .map(|c| c.len_utf8())
+                            .sum()
+                    } else {
+                        0
+                    };
+
+                    // Keep track of the changes for next offsets
+                    offset += replaced_char_size as isize;
+                    offset += total_bytes_to_remove as isize;
+
+                    new_alignments.extend((0..c.len_utf8()).map(|_| align));
+
+                    // Then we keep only the char for string reconstruction
+                    c
+                })
+                .collect::<String>();
+
+            new_normalized.push_str(&normalized);
+            last_end = end;
+        }
+
+        // Copy the remaining part of the input
+        new_normalized.push_str(&self.normalized[last_end..]);
+        new_alignments.extend(&self.alignments[last_end..]);
+
+        self.normalized = new_normalized;
+        self.alignments = new_alignments;
+        Ok(())
+    }
+
     /// Clear the normalized part of the string
     pub fn clear(&mut self) -> usize {
         let len = self.len();