diff --git a/tests/test_api.py b/tests/test_api.py
index b5f31bc3..000b94c9 100644
--- a/tests/test_api.py
+++ b/tests/test_api.py
@@ -611,3 +611,64 @@ def test_parse_accepts_nesting_at_the_depth_limit() -> None:
     assert parse(array).as_string() == array
     dotted = ".".join(["a"] * depth) + " = 1"
     assert parse(dotted).as_string() == dotted
+
+
+@pytest.mark.parametrize(
+    "raw",
+    [
+        "0",
+        "42",
+        "1_000",
+        "5_349_221",
+        "+1_000",
+        "-1_000",
+        "0xdead_beef",
+        "0xDEAD_BEEF",
+        "0o7_7",
+        "0b1_0",
+    ],
+)
+def test_integer_preserves_grouping_underscores(raw: str) -> None:
+    assert tomlkit.integer(raw).as_string() == raw
+    assert int(tomlkit.integer(raw)) == int(raw, 0)
+
+
+@pytest.mark.parametrize("raw", ["1_000", "0xdead_beef", "+1_000", "0o7_7"])
+def test_integer_emission_matches_parse(raw: str) -> None:
+    doc = tomlkit.document()
+    doc["n"] = tomlkit.integer(raw)
+    assert tomlkit.dumps(doc) == f"n = {raw}\n"
+    # the emission path agrees with the parse path on the same literal
+    assert (
+        tomlkit.integer(raw).as_string()
+        == tomlkit.parse(f"n = {raw}\n")["n"].as_string()
+    )
+
+
+@pytest.mark.parametrize(
+    "raw",
+    [
+        "",
+        "abc",
+        "_100",
+        "100_",
+        "1__000",
+        "01_2",
+        "0x_1",
+        "0X1",  # the parser only accepts a lowercase prefix
+        "+0x1",  # hex/oct/bin take no sign
+        "-0o7",
+        " 1",
+        "1 ",
+        "1.0",
+        "1e3",
+    ],
+)
+def test_integer_rejects_invalid_strings(raw: str) -> None:
+    with pytest.raises(ValueError):
+        tomlkit.integer(raw)
+
+
+def test_integer_from_int_unchanged() -> None:
+    assert tomlkit.integer(1000).as_string() == "1000"
+    assert int(tomlkit.integer(1000)) == 1000
diff --git a/tomlkit/api.py b/tomlkit/api.py
index b0f8cd6d..69fdbb53 100644
--- a/tomlkit/api.py
+++ b/tomlkit/api.py
@@ -2,6 +2,7 @@
 
 import contextlib
 import datetime as _datetime
+import re
 
 from collections.abc import Iterable
 from collections.abc import Mapping
@@ -114,8 +115,33 @@ def document() -> TOMLDocument:
 
 
 # Items
+_INTEGER_RE = re.compile(
+    r"(?:"
+    r"0x[0-9a-fA-F](?:_?[0-9a-fA-F])*"  # hex: lowercase prefix only, no sign
+    r"|0o[0-7](?:_?[0-7])*"  # octal: lowercase prefix only, no sign
+    r"|0b[01](?:_?[01])*"  # binary: lowercase prefix only, no sign
+    r"|[+-]?(?:0|[1-9](?:_?[0-9])*)"  # decimal: optional sign, no leading zeros
+    r")"
+)
+"""Valid TOML v1.0.0 integer literals (https://toml.io/en/v1.0.0#integer).
+
+Underscores may only appear between digits; hex/octal/binary use a
+lowercase prefix and take no sign, mirroring ``Parser._parse_number``.
+"""
+
+
 def integer(raw: str | int) -> Integer:
-    """Create an integer item from a number or string."""
+    """Create an integer item from a number or string.
+
+    When ``raw`` is a string it must be a valid TOML v1.0.0 integer
+    literal; the original text is then preserved verbatim so grouping
+    underscores (e.g. ``"1_000"``) survive ``dumps()``. Anything else
+    raises ``ValueError``.
+    """
+    if isinstance(raw, str):
+        if not _INTEGER_RE.fullmatch(raw):
+            raise ValueError(f"Invalid TOML integer: {raw!r}")
+        return Integer(int(raw, 0), Trivia(), raw)
     return item(int(raw))