diff --git a/pre_commit_hooks/fix_encoding_pragma.py b/pre_commit_hooks/fix_encoding_pragma.py
index bde4e78a..23fc79fd 100644
--- a/pre_commit_hooks/fix_encoding_pragma.py
+++ b/pre_commit_hooks/fix_encoding_pragma.py
@@ -9,14 +9,14 @@
 from typing import Sequence
 from typing import Union
 
-DEFAULT_PRAGMA = b'# -*- coding: utf-8 -*-\n'
+DEFAULT_PRAGMA = b'# -*- coding: utf-8 -*-'
 
 
 def has_coding(line):  # type: (bytes) -> bool
     if not line.strip():
         return False
     return (
-        line.lstrip()[0:1] == b'#' and (
+        line.lstrip()[:1] == b'#' and (
             b'unicode' in line or
             b'encoding' in line or
             b'coding:' in line or
@@ -26,7 +26,7 @@ def has_coding(line):  # type: (bytes) -> bool
 
 
 class ExpectedContents(collections.namedtuple(
-        'ExpectedContents', ('shebang', 'rest', 'pragma_status'),
+        'ExpectedContents', ('shebang', 'rest', 'pragma_status', 'ending'),
 )):
     """
     pragma_status:
@@ -47,6 +47,8 @@ def is_expected_pragma(self, remove):  # type: (bool) -> bool
 
 def _get_expected_contents(first_line, second_line, rest, expected_pragma):
     # type: (bytes, bytes, bytes, bytes) -> ExpectedContents
+    ending = b'\r\n' if first_line.endswith(b'\r\n') else b'\n'
+
     if first_line.startswith(b'#!'):
         shebang = first_line
         potential_coding = second_line
@@ -55,7 +57,7 @@ def _get_expected_contents(first_line, second_line, rest, expected_pragma):
         potential_coding = first_line
         rest = second_line + rest
 
-    if potential_coding == expected_pragma:
+    if potential_coding.rstrip(b'\r\n') == expected_pragma:
         pragma_status = True  # type: Optional[bool]
     elif has_coding(potential_coding):
         pragma_status = None
@@ -64,7 +66,7 @@ def _get_expected_contents(first_line, second_line, rest, expected_pragma):
         rest = potential_coding + rest
 
     return ExpectedContents(
-        shebang=shebang, rest=rest, pragma_status=pragma_status,
+        shebang=shebang, rest=rest, pragma_status=pragma_status, ending=ending,
     )
 
 
@@ -93,7 +95,7 @@ def fix_encoding_pragma(f, remove=False, expected_pragma=DEFAULT_PRAGMA):
     f.truncate()
     f.write(expected.shebang)
     if not remove:
-        f.write(expected_pragma)
+        f.write(expected_pragma + expected.ending)
     f.write(expected.rest)
 
     return 1
@@ -102,11 +104,7 @@ def fix_encoding_pragma(f, remove=False, expected_pragma=DEFAULT_PRAGMA):
 def _normalize_pragma(pragma):  # type: (Union[bytes, str]) -> bytes
     if not isinstance(pragma, bytes):
         pragma = pragma.encode('UTF-8')
-    return pragma.rstrip() + b'\n'
-
-
-def _to_disp(pragma):  # type: (bytes) -> str
-    return pragma.decode().rstrip()
+    return pragma.rstrip()
 
 
 def main(argv=None):  # type: (Optional[Sequence[str]]) -> int
@@ -117,7 +115,7 @@ def main(argv=None):  # type: (Optional[Sequence[str]]) -> int
     parser.add_argument(
         '--pragma', default=DEFAULT_PRAGMA, type=_normalize_pragma,
         help='The encoding pragma to use.  Default: {}'.format(
-            _to_disp(DEFAULT_PRAGMA),
+            DEFAULT_PRAGMA.decode(),
         ),
     )
     parser.add_argument(
@@ -141,7 +139,7 @@ def main(argv=None):  # type: (Optional[Sequence[str]]) -> int
             retv |= file_ret
             if file_ret:
                 print(fmt.format(
-                    pragma=_to_disp(args.pragma), filename=filename,
+                    pragma=args.pragma.decode(), filename=filename,
                 ))
 
     return retv
diff --git a/tests/fix_encoding_pragma_test.py b/tests/fix_encoding_pragma_test.py
index 7288bfa1..d94b7256 100644
--- a/tests/fix_encoding_pragma_test.py
+++ b/tests/fix_encoding_pragma_test.py
@@ -112,7 +112,7 @@ def test_not_ok_inputs(input_str, output):
 def test_ok_input_alternate_pragma():
     input_s = b'# coding: utf-8\nx = 1\n'
     bytesio = io.BytesIO(input_s)
-    ret = fix_encoding_pragma(bytesio, expected_pragma=b'# coding: utf-8\n')
+    ret = fix_encoding_pragma(bytesio, expected_pragma=b'# coding: utf-8')
     assert ret == 0
     bytesio.seek(0)
     assert bytesio.read() == input_s
@@ -120,7 +120,7 @@ def test_ok_input_alternate_pragma():
 
 def test_not_ok_input_alternate_pragma():
     bytesio = io.BytesIO(b'x = 1\n')
-    ret = fix_encoding_pragma(bytesio, expected_pragma=b'# coding: utf-8\n')
+    ret = fix_encoding_pragma(bytesio, expected_pragma=b'# coding: utf-8')
     assert ret == 1
     bytesio.seek(0)
     assert bytesio.read() == b'# coding: utf-8\nx = 1\n'
@@ -130,11 +130,11 @@ def test_not_ok_input_alternate_pragma():
     ('input_s', 'expected'),
     (
         # Python 2 cli parameters are bytes
-        (b'# coding: utf-8', b'# coding: utf-8\n'),
+        (b'# coding: utf-8', b'# coding: utf-8'),
         # Python 3 cli parameters are text
-        ('# coding: utf-8', b'# coding: utf-8\n'),
+        ('# coding: utf-8', b'# coding: utf-8'),
         # trailing whitespace
-        ('# coding: utf-8\n', b'# coding: utf-8\n'),
+        ('# coding: utf-8\n', b'# coding: utf-8'),
     ),
 )
 def test_normalize_pragma(input_s, expected):
@@ -150,3 +150,16 @@ def test_integration_alternate_pragma(tmpdir, capsys):
     assert f.read() == '# coding: utf-8\nx = 1\n'
     out, _ = capsys.readouterr()
     assert out == 'Added `# coding: utf-8` to {}\n'.format(f.strpath)
+
+
+def test_crlf_ok(tmpdir):
+    f = tmpdir.join('f.py')
+    f.write_binary(b'# -*- coding: utf-8 -*-\r\nx = 1\r\n')
+    assert not main((f.strpath,))
+
+
+def test_crfl_adds(tmpdir):
+    f = tmpdir.join('f.py')
+    f.write_binary(b'x = 1\r\n')
+    assert main((f.strpath,))
+    assert f.read_binary() == b'# -*- coding: utf-8 -*-\r\nx = 1\r\n'