diff --git a/pristine_zip/delta_to_zipball.py b/pristine_zip/delta_to_zipball.py
index 5255dcc..96c7d7f 100644
--- a/pristine_zip/delta_to_zipball.py
+++ b/pristine_zip/delta_to_zipball.py
@@ -1,46 +1,61 @@
 # Copyright (C) 2020 The Software Heritage developers
 # See the AUTHORS file at the top-level directory of this distribution
 # License: GNU General Public License version 3, or any later version
 # See top-level LICENSE file for more information
 
 import os
 import subprocess
 import tempfile
 
 from . import compress
 
 
 def genzip(checkout_dir: str, delta_path: str, zipball_path: str):
     with tempfile.TemporaryDirectory(prefix="pristine-zip-genzip") as work_dir:
         reference_zipball_path = os.path.join(work_dir, "reference.zip")
 
         # generate reference zipball
         compress.compress(checkout_dir, reference_zipball_path)
 
         _apply_delta(reference_zipball_path, zipball_path, work_dir, delta_path)
 
 
 def _apply_delta(
     reference_zipball_path: str, zipball_path: str, work_dir: str, delta_path: str,
 ):
     proc = subprocess.run(["tar", "--extract", "-f", delta_path,], cwd=work_dir)
     proc.check_returncode()
 
     with open(os.path.join(work_dir, "type"), "rb") as fd:
         type_ = fd.read().decode().strip()
         assert type_ == "zip", (
             f"Unknown zipball type {type}. Are you"
             f"extracting a delta from pristine-tar instead of pristine-zip?"
         )
 
+    with open(os.path.join(work_dir, "reference_md5sum"), "rb") as fd:
+        expected_md5sum = fd.read().decode().strip()
+
+    proc = subprocess.run(["md5sum", reference_zipball_path], capture_output=True)
+    proc.check_returncode()
+    actual_md5sum = proc.stdout.decode().strip()
+
+    if actual_md5sum != expected_md5sum:
+        print(
+            f"md5sum mismatch between reference zipballs. "
+            f"This is a bug, please report it along with the original zipball "
+            f"and the version number of pristine-zip."
+        )
+        exit(1)
+
     xdelta3_path = os.path.join(work_dir, "delta")
     assert os.path.isfile(xdelta3_path), "Missing 'delta' file in delta archive."
 
     _apply_xdelta3(reference_zipball_path, zipball_path, xdelta3_path)
 
 
 def _apply_xdelta3(reference_zipball_path: str, zipball_path: str, xdelta3_path: str):
     proc = subprocess.run(
         ["xdelta3", "-d", "-s", reference_zipball_path, xdelta3_path, zipball_path]
     )
     proc.check_returncode()
diff --git a/pristine_zip/zipball_to_delta.py b/pristine_zip/zipball_to_delta.py
index 3bdfd07..0104929 100644
--- a/pristine_zip/zipball_to_delta.py
+++ b/pristine_zip/zipball_to_delta.py
@@ -1,90 +1,97 @@
 # Copyright (C) 2020 The Software Heritage developers
 # See the AUTHORS file at the top-level directory of this distribution
 # License: GNU General Public License version 3, or any later version
 # See top-level LICENSE file for more information
 
 import os
 import subprocess
 import tempfile
 
 from . import compress
 
 
 def gendelta(zipball_path: str, delta_path: str):
     with tempfile.TemporaryDirectory(prefix="pristine-zip-gendelta") as work_dir:
         reference_zipball_path = os.path.join(work_dir, "reference.zip")
         generate_reference_zipball_from_zipball(
             zipball_path, reference_zipball_path, work_dir
         )
         _generate_delta(zipball_path, reference_zipball_path, work_dir, delta_path)
 
 
 def _generate_delta(
     upstream_zipball_path: str,
     reference_zipball_path: str,
     work_dir: str,
     delta_path: str,
 ):
     _generate_xdelta3(
         upstream_zipball_path, reference_zipball_path, os.path.join(work_dir, "delta")
     )
     with open(os.path.join(work_dir, "type"), "wb") as fd:
         fd.write(b"zip\n")
 
-    files = ["delta", "type"]
+    proc = subprocess.run(["md5sum", reference_zipball_path], capture_output=True)
+    proc.check_returncode()
+    md5sum = proc.stdout.split(b" ", 1)[0]
+
+    with open(os.path.join(work_dir, "reference_md5sum"), "wb") as fd:
+        fd.write(md5sum + b"\n")
+
+    files = ["delta", "reference_md5sum", "type"]
 
     # Make the timestamps in the delta tarball deterministic
     for file in files:
         os.utime(os.path.join(work_dir, file), times=(0, 0))
 
     proc = subprocess.run(
         [
             "tar",
             # make entries in the delta tarball deterministic:
             "--owner",
             "0",
             "--group",
             "0",
             "--numeric-owner",
             "--mode",
             "644",
             # generic options:
             "--create",
             "--compress",
             "-f",
             delta_path,
             *files,
         ],
         cwd=work_dir,
     )
     proc.check_returncode()
 
 
 def _generate_xdelta3(
     upstream_zipball_path: str, reference_zipball_path: str, delta_path: str
 ):
     """Generates the xdelta3 difference between a reference zipball and
     the original one."""
     proc = subprocess.run(
         [
             "xdelta3",
             "-e",
             "-s",
             reference_zipball_path,
             upstream_zipball_path,
             delta_path,
         ]
     )
     proc.check_returncode()
 
 
 def generate_reference_zipball_from_zipball(
     upstream_zipball_path: str, reference_zipball_path: str, work_dir: str
 ):
     """Unzips an upstream zipball and rezips it in a reference zipball."""
     checkout_dir = os.path.join(work_dir, "checkout")
     os.mkdir(checkout_dir)
     proc = subprocess.run(["unzip", upstream_zipball_path], cwd=checkout_dir)
     proc.check_returncode()
 
     compress.compress(checkout_dir, reference_zipball_path)