diff --git a/scrapy/core/downloader/handlers/s3.py b/scrapy/core/downloader/handlers/s3.py
index 6d28f866e..0903b84ad 100644
--- a/scrapy/core/downloader/handlers/s3.py
+++ b/scrapy/core/downloader/handlers/s3.py
@@ -85,15 +85,10 @@ class S3DownloadHandler(object):
             request = request.replace(url=url)
         elif self._signer is not None:
             import botocore.awsrequest
-            from botocore.vendored.requests.structures import CaseInsensitiveDict
-            print(url, request.headers)
             awsrequest = botocore.awsrequest.AWSRequest(
                 method=request.method,
                 url='%s://s3.amazonaws.com/%s%s' % (scheme, bucket, path),
-                # TODO - move to a header method
-                headers=CaseInsensitiveDict(
-                    (to_unicode(key), to_unicode(b','.join(value)))
-                    for key, value in request.headers.items()),
+                headers=request.headers.to_native_string_dict(),
                 data=request.body)
             self._signer.add_auth(awsrequest)
             request = request.replace(
diff --git a/scrapy/http/headers.py b/scrapy/http/headers.py
index 13f0f0383..d0c4cd0fb 100644
--- a/scrapy/http/headers.py
+++ b/scrapy/http/headers.py
@@ -1,6 +1,7 @@
 import six
 from w3lib.http import headers_dict_to_raw
 from scrapy.utils.datatypes import CaselessDict
+from scrapy.utils.python import to_unicode
 
 
 class Headers(CaselessDict):
@@ -78,6 +79,12 @@ class Headers(CaselessDict):
     def to_string(self):
         return headers_dict_to_raw(self)
 
+    def to_native_string_dict(self):
+        return CaselessDict(
+            (to_unicode(key, encoding=self.encoding),
+             to_unicode(b','.join(value), encoding=self.encoding))
+            for key, value in self.items())
+
     def __copy__(self):
         return self.__class__(self)
     copy = __copy__