diff --git a/ofxtools/header.py b/ofxtools/header.py
index 1f7bbba..9d26c8b 100644
--- a/ofxtools/header.py
+++ b/ofxtools/header.py
@@ -261,14 +261,26 @@ def parse_header(source: BinaryIO) -> Tuple[OFXHeaderType, str]:
     """
     logger.info("Parsing OFX header")
 
-    # Skip any empty lines at the beginning
-    while True:
+    # Skip empty lines at the beginning
+    # If we can't find any content in the first 8 lines, there's
+    # something wrong with the data.
+    found_header = False
+    for _ in range(8):
+        # Remember the position within the file where the header begins
+        # We'll need this, plus the offset to the end of the regex, to seek() to the
+        # start of the body tag soup when parsing OFXv1
+        header_start = source.tell()
+
         # OFX header is read by nice clean machines, not meatbags -
         # should not contain 💩, 漢字, or what have you.
         line = source.readline().decode("ascii")
         if line.strip():
+            found_header = True
             break
 
+    if not found_header:
+        raise OFXHeaderError(f"Invalid OFX header - {source}")
+
     # If the first non-empty line contains an XML declaration, it's OFX v2
     xml_match = XML_REGEX.match(line)
     if xml_match:
@@ -289,9 +301,10 @@ def parse_header(source: BinaryIO) -> Tuple[OFXHeaderType, str]:
         rawheader = line + "\n"
         # First line is OFXHEADER; need to read next 8 lines for a fixed
         # total of 9 fields required by OFX v1 spec.
-        for n in range(8):
+        for _ in range(8):
             rawheader += source.readline().decode("ascii")
-        header, header_end_index = OFXHeaderV1.parse(rawheader)
+
+        header, header_end_offset = OFXHeaderV1.parse(rawheader)
 
         #  Input source stream position should have advanced to the beginning of
         #  the OFX body tag soup, which is where subsequent calls
@@ -299,13 +312,13 @@ def parse_header(source: BinaryIO) -> Tuple[OFXHeaderType, str]:
         #
         #  The seek call will correct the position when \r newline character is used
         #  (Issue #84)
-        source.seek(header_end_index)
+        source.seek(header_start + header_end_offset)
 
         #  Decode the OFX data body according to the encoding declared
         #  in the OFX header
-        message = source.read().decode(header.codec)
+        message = source.read().decode(header.codec).strip()
 
-    return header, message.strip()
+    return header, message
 
 
 def make_header(
diff --git a/tests/test_header.py b/tests/test_header.py
index 3c9392b..432bf12 100644
--- a/tests/test_header.py
+++ b/tests/test_header.py
@@ -390,6 +390,46 @@ def testNoLineBreaksAnywhere(self):
 
         self.assertEqual(body, body_)
 
+    def testHeaderlessDoomScroll(self):
+        # Some FIs apparently send endless empty lines with no header.
+        # To prevent that, we allow a maximum of 7 empty lines
+        # before getting down to brass tacks.
+        header = "\r\n" * 7
+        header += (
+            "OFXHEADER:  100\r\n"
+            "DATA:  OFXSGML\r\n"
+            "VERSION:  160\r\n"
+            "SECURITY:  NONE\r\n"
+            "ENCODING:  USASCII\r\n"
+            "CHARSET:  NONE\r\n"
+            "COMPRESSION:  NONE\r\n"
+            "OLDFILEUID:  NONE\r\n"
+            "NEWFILEUID:  NONE\r\n"
+        )
+
+        ofx = header + self.body
+        ofx = BytesIO(ofx.encode("utf8"))
+        ofxheader, body = ofxtools.header.parse_header(ofx)
+
+        self.assertEqual(ofxheader.ofxheader, 100)
+        self.assertEqual(ofxheader.data, "OFXSGML")
+        self.assertEqual(ofxheader.version, 160)
+        self.assertEqual(ofxheader.security, "NONE")
+        self.assertEqual(ofxheader.encoding, "USASCII")
+        self.assertEqual(ofxheader.charset, "NONE")
+        self.assertEqual(ofxheader.compression, "NONE")
+        self.assertEqual(ofxheader.oldfileuid, "NONE")
+        self.assertEqual(ofxheader.newfileuid, "NONE")
+
+        self.assertEqual(body, self.body)
+
+        # 8 empty lines before header should fail
+        header = "\r\n" + header
+        ofx = header + self.body
+        ofx = BytesIO(ofx.encode("utf8"))
+        with self.assertRaises(ofxtools.header.OFXHeaderError):
+            ofxtools.header.parse_header(ofx)
+
 
 class OFXHeaderV2TestCase(unittest.TestCase, OFXHeaderTestMixin):
     headerClass = ofxtools.header.OFXHeaderV2