diff --git a/repomirror/utils/find_fastest_upstream/__init__.py b/repomirror/utils/find_fastest_upstream/__init__.py
index e69de29..ae9c9f3 100644
--- a/repomirror/utils/find_fastest_upstream/__init__.py
+++ b/repomirror/utils/find_fastest_upstream/__init__.py
@@ -0,0 +1,2 @@
+from . import constants
+from . import classes
diff --git a/repomirror/utils/find_fastest_upstream/centos.py b/repomirror/utils/find_fastest_upstream/centos.py
index 973a6dc..af496b7 100755
--- a/repomirror/utils/find_fastest_upstream/centos.py
+++ b/repomirror/utils/find_fastest_upstream/centos.py
@@ -2,7 +2,6 @@
 
 import argparse
 import csv
-import datetime
 import io
 import re
 ##
diff --git a/repomirror/utils/find_fastest_upstream/centos_epel.py b/repomirror/utils/find_fastest_upstream/centos_epel.py
new file mode 100755
index 0000000..e2ed9fa
--- /dev/null
+++ b/repomirror/utils/find_fastest_upstream/centos_epel.py
@@ -0,0 +1,94 @@
+#!/usr/bin/env python3
+
+import argparse
+import re
+##
+import classes
+
+
+_proto_re = re.compile(r'^(?P<proto>https?)(?P<uri>.*)')
+
+
+class Ranker(classes.Ranker):
+    # No CSV, JSON, or XML that I could find, unfortunately.
+    # There's apparently? an API to mirrormanager2 but I can't seem to find a public endpoint nor an endpoint that
+    # would return the mirrors.
+    mirrorlist_url = 'https://admin.fedoraproject.org/mirrormanager/mirrors/EPEL'
+    distro_name = 'EPEL'
+
+    def __init__(self, *args, **kwargs):
+        super().__init__(*args, **kwargs)
+        self.get_mirrors()
+
+    def extract_mirrors(self, preferred_proto = 'rsync'):
+        preferred_proto = preferred_proto.lower()
+        if preferred_proto not in ('rsync', 'ftp'):
+            raise ValueError('Invalid preferred_proto; must be one of rsync or ftp')
+        non_preferred = ('rsync' if preferred_proto == 'ftp' else 'ftp')
+        mirror_section = self.bs.find('h2', string = 'Public active mirrors')
+        mirror_table = mirror_section.find_next('table')
+        if mirror_table is None:
+            return(None)
+        # https://stackoverflow.com/a/56835562/733214
+        headers = [h.text for h in mirror_table.find_all('th')]
+        rows = [m for m in mirror_table.find_all('tr')][1:]
+        for row in rows:
+            mirror = {}
+            do_skip = False
+            for idx, cell in enumerate(row.find_all('td')):
+                k = headers[idx]
+                v = cell.text.strip()
+                if k == 'Country' and v != self.my_info['country']:
+                    do_skip = True
+                    continue
+                if k == 'Categories' and not do_skip:
+                    if 'EPEL' not in v:
+                        do_skip = True
+                        continue
+                    pref_proto = cell.find('a', attrs = {
+                        'href': re.compile(r'^{0}://'.format(preferred_proto), re.IGNORECASE)})
+                    non_pref = cell.find('a', attrs = {
+                        'href': re.compile(r'^{0}://'.format(non_preferred), re.IGNORECASE)})
+                    if pref_proto is not None:
+                        v = pref_proto['href']
+                    elif non_pref is not None:
+                        v = non_pref['href']
+                    else:
+                        v = None
+                    mirror['url'] = v
+                # Fedora team can't spell.
+                elif k in ('Bandwidth', 'Bandwith'):
+                    mirror['bw'] = int(v)
+            if do_skip:
+                continue
+            if not mirror['url']:
+                continue
+            self.raw_mirrors.append(mirror)
+            self.mirror_candidates.append(mirror['url'])
+        return(None)
+
+
+def parseArgs():
+    args = argparse.ArgumentParser(description = 'Generate a list of suitable EPEL upstream mirrors in order of '
+                                                 'speed.')
+    args.add_argument('-x', '--xml',
+                      dest = 'xml',
+                      action = 'store_true',
+                      help = ('If specified, generate a config stub instead of a printed list of URLs'))
+    return(args)
+
+
+def main():
+    args = parseArgs().parse_args()
+    r = Ranker()
+    r.extract_mirrors()
+    r.speedcheck()
+    if args.xml:
+        print(r.gen_xml())
+    else:
+        r.print()
+    return(None)
+
+
+if __name__ == '__main__':
+    main()