diff --git a/workflow_configuration/fix_invalid_readingorder.py b/workflow_configuration/fix_invalid_readingorder.py
new file mode 100644
index 0000000..f9e1a1a
--- /dev/null
+++ b/workflow_configuration/fix_invalid_readingorder.py
@@ -0,0 +1,45 @@
+import click
+import shutil
+import sys
+
+from lxml import etree
+
+from ocrd_models.constants import NAMESPACES
+from ocrd_models.utils import xmllint_format
+from ocrd_validators import PageValidator
+from ocrd_validators.page_validator import ReadingOrderInvalidError
+
+
+@click.command(context_settings=dict(help_option_names=['-h', '--help']))
+@click.argument('pagefile', type=click.Path(exists=True, dir_okay=False))
+@click.option('-I', '--inplace', is_flag=True, default=False, help="Write result back to input file intead of stdout")
+@click.option('-B', '--backup', default='', help="Write a copy of the original file under this suffix")
+def cli(pagefile, inplace, backup):
+    """Remove invalid references from the ReadingOrder."""
+
+    report = PageValidator.validate(filename=pagefile)
+    if report.is_valid:
+        print("Nothing to do, page validation found no ReadingOrder issues.", file=sys.stderr)
+        return
+
+    with click.open_file(pagefile, 'rb') as page:
+        tree = etree.parse(page)
+
+    for err in report.errors:
+        if isinstance(err, ReadingOrderInvalidError):
+            ref = tree.find(f'//*[@regionRef="{err.region_ref}"]', namespaces=NAMESPACES)
+            ref.getparent().remove(ref)
+    
+    result = etree.tostring(etree.ElementTree(tree.getroot()), pretty_print=True, encoding='UTF-8')
+    result = xmllint_format(result)
+    
+    if inplace:
+        if backup:
+            shutil.copyfile(pagefile, pagefile + backup)
+        with click.open_file(pagefile, 'wb') as page:
+            page.write(result)
+    else:
+        sys.stdout.buffer.write(result)
+
+if __name__ == '__main__':
+    cli()