diff --git a/swh/model/hypothesis_strategies.py b/swh/model/hypothesis_strategies.py
--- a/swh/model/hypothesis_strategies.py
+++ b/swh/model/hypothesis_strategies.py
@@ -18,6 +18,7 @@
     from_regex,
     integers,
     just,
+    lists,
     none,
     one_of,
     sampled_from,
@@ -229,6 +230,12 @@
 revision_metadata = metadata_dicts
 
 
+def extra_headers():
+    return lists(
+        tuples(binary(min_size=0, max_size=50), binary(min_size=0, max_size=500))
+    ).map(tuple)
+
+
 def revisions_d():
     return builds(
         dict,
@@ -242,6 +249,7 @@
         directory=sha1_git(),
         type=sampled_from([x.value for x in RevisionType]),
         metadata=optional(revision_metadata()),
+        extra_headers=extra_headers(),
     )
     # TODO: metadata['extra_headers'] can have binary keys and values
 
diff --git a/swh/model/identifiers.py b/swh/model/identifiers.py
--- a/swh/model/identifiers.py
+++ b/swh/model/identifiers.py
@@ -426,7 +426,7 @@
     - author_date
     - committer
     - committer_date
-    - metadata -> extra_headers
+    - extra_headers or metadata -> extra_headers
     - message
 
     A revision's identifier is the 'git'-checksum of a commit manifest
@@ -486,22 +486,13 @@
     )
 
     # Handle extra headers
-    metadata = revision.get("metadata")
-    if not metadata:
-        metadata = {}
+    metadata = revision.get("metadata") or {}
+    extra_headers = revision.get("extra_headers", ())
+    if not extra_headers and "extra_headers" in metadata:
+        extra_headers = metadata["extra_headers"]
 
-    for key, value in metadata.get("extra_headers", []):
-
-        # Integer values: decimal representation
-        if isinstance(value, int):
-            value = str(value).encode("utf-8")
-
-        # Unicode string values: utf-8 encoding
-        if isinstance(value, str):
-            value = value.encode("utf-8")
-
-        # encode the key to utf-8
-        components.extend([key.encode("utf-8"), b" ", escape_newlines(value), b"\n"])
+    for key, value in extra_headers:
+        components.extend([key, b" ", escape_newlines(value), b"\n"])
 
     if revision["message"] is not None:
         components.extend([b"\n", revision["message"]])
diff --git a/swh/model/model.py b/swh/model/model.py
--- a/swh/model/model.py
+++ b/swh/model/model.py
@@ -6,9 +6,10 @@
 import datetime
 
 from abc import ABCMeta, abstractmethod
+from copy import deepcopy
 from enum import Enum
 from hashlib import sha256
-from typing import Dict, Optional, Tuple, TypeVar, Union
+from typing import Dict, Iterable, Optional, Tuple, TypeVar, Union
 from typing_extensions import Final
 
 import attr
@@ -410,6 +411,10 @@
     MERCURIAL = "hg"
 
 
+def tuplify_extra_headers(value: Iterable) -> Tuple:
+    return tuple((k, v) for k, v in value)
+
+
 @attr.s(frozen=True)
 class Revision(BaseModel, HashableObject):
     object_type: Final = "revision"
@@ -429,6 +434,27 @@
     )
     parents = attr.ib(type=Tuple[Sha1Git, ...], validator=type_validator(), default=())
     id = attr.ib(type=Sha1Git, validator=type_validator(), default=b"")
+    extra_headers = attr.ib(
+        type=Tuple[Tuple[bytes, bytes], ...],  # but it makes mypy sad
+        validator=type_validator(),
+        converter=tuplify_extra_headers,  # type: ignore
+        default=(),
+    )
+
+    def __attrs_post_init__(self):
+        super().__attrs_post_init__()
+        # ensure metadata is a deep copy of whatever was given, and if needed
+        # extract extra_headers from there
+        if self.metadata:
+            metadata = deepcopy(self.metadata)
+            if not self.extra_headers and "extra_headers" in metadata:
+                object.__setattr__(
+                    self,
+                    "extra_headers",
+                    tuplify_extra_headers(metadata.pop("extra_headers")),
+                )
+                attr.validate(self)
+            object.__setattr__(self, "metadata", metadata)
 
     @staticmethod
     def compute_hash(object_dict):
diff --git a/swh/model/tests/test_identifiers.py b/swh/model/tests/test_identifiers.py
--- a/swh/model/tests/test_identifiers.py
+++ b/swh/model/tests/test_identifiers.py
@@ -370,12 +370,10 @@
                 2015, 7, 12, 15, 10, 30, tzinfo=linus_tz
             ),
             "message": b"Linux 4.2-rc2\n",
-            "metadata": {
-                "extra_headers": [
-                    ["svn-repo-uuid", "046f1af7-66c2-d61b-5410-ce57b7db7bff"],
-                    ["svn-revision", 10],
-                ]
-            },
+            "extra_headers": (
+                (b"svn-repo-uuid", b"046f1af7-66c2-d61b-5410-ce57b7db7bff"),
+                (b"svn-revision", b"10"),
+            ),
         }
 
         self.revision_with_gpgsig = {
@@ -393,7 +391,7 @@
             "date": {"timestamp": 1428538899, "offset": 480,},
             "committer": {"name": b"Jiang Xin", "email": b"worldhello.net@gmail.com",},
             "committer_date": {"timestamp": 1428538899, "offset": 480,},
-            "metadata": {"extra_headers": [["gpgsig", gpgsig],],},
+            "extra_headers": ((b"gpgsig", gpgsig),),
             "message": b"""Merge branch 'master' of git://github.com/alexhenrie/git-po
 
 * 'master' of git://github.com/alexhenrie/git-po:
@@ -450,12 +448,10 @@
                 2015, 7, 12, 15, 10, 30, tzinfo=linus_tz
             ),
             "message": b"Linux 4.2-rc2\n",
-            "metadata": {
-                "extra_headers": [
-                    ["svn-repo-uuid", "046f1af7-66c2-d61b-5410-ce57b7db7bff"],
-                    ["svn-revision", 10],
-                ]
-            },
+            "extra_headers": (
+                (b"svn-repo-uuid", b"046f1af7-66c2-d61b-5410-ce57b7db7bff"),
+                (b"svn-revision", b"10"),
+            ),
         }
 
     def test_revision_identifier(self):
diff --git a/swh/model/tests/test_model.py b/swh/model/tests/test_model.py
--- a/swh/model/tests/test_model.py
+++ b/swh/model/tests/test_model.py
@@ -415,6 +415,194 @@
         SkippedContent.from_dict(skipped_content_d)
 
 
+# Revision
+
+
+def test_revision_extra_headers_no_headers():
+    rev_dict = revision_example.copy()
+    rev_dict.pop("id")
+    rev = Revision.from_dict(rev_dict)
+    rev_dict = attr.asdict(rev, recurse=False)
+
+    rev_model = Revision(**rev_dict)
+    assert rev_model.metadata is None
+    assert rev_model.extra_headers == ()
+
+    rev_dict["metadata"] = {
+        "something": "somewhere",
+        "some other thing": "stranger",
+    }
+    rev_model = Revision(**rev_dict)
+    assert rev_model.metadata == rev_dict["metadata"]
+    assert rev_model.extra_headers == ()
+
+
+def test_revision_extra_headers_with_headers():
+    rev_dict = revision_example.copy()
+    rev_dict.pop("id")
+    rev = Revision.from_dict(rev_dict)
+    rev_dict = attr.asdict(rev, recurse=False)
+    rev_dict["metadata"] = {
+        "something": "somewhere",
+        "some other thing": "stranger",
+    }
+    extra_headers = (
+        (b"header1", b"value1"),
+        (b"header2", b"42"),
+        (b"header3", b"should I?\u0000"),
+        (b"header1", b"again"),
+    )
+
+    rev_dict["extra_headers"] = extra_headers
+    rev_model = Revision(**rev_dict)
+    assert "extra_headers" not in rev_model.metadata
+    assert rev_model.extra_headers == extra_headers
+
+
+def test_revision_extra_headers_in_metadata():
+    rev_dict = revision_example.copy()
+    rev_dict.pop("id")
+    rev = Revision.from_dict(rev_dict)
+    rev_dict = attr.asdict(rev, recurse=False)
+    rev_dict["metadata"] = {
+        "something": "somewhere",
+        "some other thing": "stranger",
+    }
+
+    extra_headers = (
+        (b"header1", b"value1"),
+        (b"header2", b"42"),
+        (b"header3", b"should I?\u0000"),
+        (b"header1", b"again"),
+    )
+
+    # check the bw-compat init hook does the job
+    # ie. extra_headers are given in the metadata field
+    rev_dict["metadata"]["extra_headers"] = extra_headers
+    rev_model = Revision(**rev_dict)
+    assert "extra_headers" not in rev_model.metadata
+    assert rev_model.extra_headers == extra_headers
+
+
+def test_revision_extra_headers_as_lists():
+    rev_dict = revision_example.copy()
+    rev_dict.pop("id")
+    rev = Revision.from_dict(rev_dict)
+    rev_dict = attr.asdict(rev, recurse=False)
+    rev_dict["metadata"] = {}
+
+    extra_headers = (
+        (b"header1", b"value1"),
+        (b"header2", b"42"),
+        (b"header3", b"should I?\u0000"),
+        (b"header1", b"again"),
+    )
+
+    # check Revision.extra_headers tuplify does the job
+    rev_dict["extra_headers"] = [list(x) for x in extra_headers]
+    rev_model = Revision(**rev_dict)
+    assert "extra_headers" not in rev_model.metadata
+    assert rev_model.extra_headers == extra_headers
+
+
+def test_revision_extra_headers_type_error():
+    rev_dict = revision_example.copy()
+    rev_dict.pop("id")
+    rev = Revision.from_dict(rev_dict)
+    orig_rev_dict = attr.asdict(rev, recurse=False)
+    orig_rev_dict["metadata"] = {
+        "something": "somewhere",
+        "some other thing": "stranger",
+    }
+    extra_headers = (
+        ("header1", b"value1"),
+        (b"header2", 42),
+        ("header1", "again"),
+    )
+    # check headers one at a time
+    #   if given as extra_header
+    for extra_header in extra_headers:
+        rev_dict = copy.deepcopy(orig_rev_dict)
+        rev_dict["extra_headers"] = (extra_header,)
+        with pytest.raises(AttributeTypeError):
+            Revision(**rev_dict)
+    #   if given as metadata
+    for extra_header in extra_headers:
+        rev_dict = copy.deepcopy(orig_rev_dict)
+        rev_dict["metadata"]["extra_headers"] = (extra_header,)
+        with pytest.raises(AttributeTypeError):
+            Revision(**rev_dict)
+
+
+def test_revision_extra_headers_from_dict():
+    rev_dict = revision_example.copy()
+    rev_dict.pop("id")
+    rev_model = Revision.from_dict(rev_dict)
+    assert rev_model.metadata is None
+    assert rev_model.extra_headers == ()
+
+    rev_dict["metadata"] = {
+        "something": "somewhere",
+        "some other thing": "stranger",
+    }
+    rev_model = Revision.from_dict(rev_dict)
+    assert rev_model.metadata == rev_dict["metadata"]
+    assert rev_model.extra_headers == ()
+
+    extra_headers = (
+        (b"header1", b"value1"),
+        (b"header2", b"42"),
+        (b"header3", b"should I?\nmaybe\x00\xff"),
+        (b"header1", b"again"),
+    )
+    rev_dict["extra_headers"] = extra_headers
+    rev_model = Revision.from_dict(rev_dict)
+    assert "extra_headers" not in rev_model.metadata
+    assert rev_model.extra_headers == extra_headers
+
+
+def test_revision_extra_headers_in_metadata_from_dict():
+    rev_dict = revision_example.copy()
+    rev_dict.pop("id")
+
+    rev_dict["metadata"] = {
+        "something": "somewhere",
+        "some other thing": "stranger",
+    }
+    extra_headers = (
+        (b"header1", b"value1"),
+        (b"header2", b"42"),
+        (b"header3", b"should I?\nmaybe\x00\xff"),
+        (b"header1", b"again"),
+    )
+    # check the bw-compat init hook does the job
+    rev_dict["metadata"]["extra_headers"] = extra_headers
+    rev_model = Revision.from_dict(rev_dict)
+    assert "extra_headers" not in rev_model.metadata
+    assert rev_model.extra_headers == extra_headers
+
+
+def test_revision_extra_headers_as_lists_from_dict():
+    rev_dict = revision_example.copy()
+    rev_dict.pop("id")
+    rev_model = Revision.from_dict(rev_dict)
+    rev_dict["metadata"] = {
+        "something": "somewhere",
+        "some other thing": "stranger",
+    }
+    extra_headers = (
+        (b"header1", b"value1"),
+        (b"header2", b"42"),
+        (b"header3", b"should I?\nmaybe\x00\xff"),
+        (b"header1", b"again"),
+    )
+    # check Revision.extra_headers converter does the job
+    rev_dict["extra_headers"] = [list(x) for x in extra_headers]
+    rev_model = Revision.from_dict(rev_dict)
+    assert "extra_headers" not in rev_model.metadata
+    assert rev_model.extra_headers == extra_headers
+
+
 # ID computation