diff --git a/tests/test_urls.py b/tests/test_urls.py
index 6e3e1f0..1c08b67 100644
--- a/tests/test_urls.py
+++ b/tests/test_urls.py
@@ -1,58 +1,83 @@
 
+import time
+from pathlib import Path
+
 import requests
 
 from openmc_data import all_release_details, all_h5_release_details, all_chain_release_details
 
 
+# urls that serve a file kept in this repository, see check_url
+REPO_RAW_PREFIX = "https://github.com/openmc-data-storage/openmc_data/raw/main/"
+REPO_ROOT = Path(__file__).resolve().parent.parent
+
+
+def check_url(url, attempts=4):
+    """Asserts that a url is usable, retrying transient failures.
+
+    Files that are stored in this repository are checked on disk rather than
+    over HTTP. A file added by a pull request is not on main until that branch
+    is merged, so fetching it would fail for exactly the change that adds it.
+    Looking for the file on disk tests the same thing, that the url in the
+    registry corresponds to a real file.
+
+    Every other url is fetched. GitHub and some of the data hosts answer 429 or
+    an occasional 404 when this suite requests hundreds of files in one job, so
+    a failure is retried with a growing wait before it is treated as real.
+    """
+    if url.startswith(REPO_RAW_PREFIX):
+        local_path = REPO_ROOT / url[len(REPO_RAW_PREFIX):]
+        assert local_path.is_file(), f"{url} points at {local_path} which is missing"
+        return
+
+    status_code = None
+    for attempt in range(attempts):
+        if attempt:
+            time.sleep(2**attempt)
+        status_code = requests.get(url, stream=True).status_code
+        if status_code == 200:
+            return
+    assert status_code == 200, (
+        f"{url} returned {status_code} on {attempts} attempts"
+    )
+
+
 def test_convert_urls():
     """Cycles through all the urls in each nuclear data library and checks
     that they return a status 200 code (success)"""
 
-    print("library, release, particle, response.status_code")
     for library, releases in all_release_details.items():
         for release, particles in releases.items():
             for particle, file_types in particles.items():
                 for file_types, value in file_types.items():
                     for file in value['compressed_files']:
                         url = value['base_url'] + file
-                        print(library, release, particle, url)
-                        response = requests.get(url, stream=True)
-                        print(library, release, particle, url, response.status_code)
                         # printing output so that in the event of a failure the
                         # failing url can be identified
-                        assert response.status_code == 200
+                        print(library, release, particle, url)
+                        check_url(url)
 
 
 def test_h5_urls():
     """Cycles through all the h5 urls in each nuclear data library and checks
     that they return a status 200 code (success)"""
 
-    print("library, release, particle, response.status_code")
     for library, releases in all_h5_release_details.items():
         for release, particles in releases.items():
             for particle, value in particles.items():
                 for file in value['compressed_files']:
                     url = value['base_url'] + file
                     print(library, release, particle, url)
-                    response = requests.get(url, stream=True)
-                    print(library, release, particle, url, response.status_code)
-                    # printing output so that in the event of a failure the
-                    # failing url can be identified
-                    assert response.status_code == 200
+                    check_url(url)
 
 
 def test_xml_urls():
     """Cycles through all the xml urls in each nuclear data library and checks
     that they return a status 200 code (success)"""
 
-    print("library, release, particle, response.status_code")
     for library, releases in all_chain_release_details.items():
         for release, branching_ratios in releases.items():
             for branching_ratio in branching_ratios:
                 url = all_chain_release_details[library][release][branching_ratio]['chain']['url']
                 print(library, release, branching_ratio, url)
-                response = requests.get(url, stream=True)
-                print(library, release, url, branching_ratio, response.status_code)
-                # printing output so that in the event of a failure the
-                # failing url can be identified
-                assert response.status_code == 200
+                check_url(url)