diff --git a/scrapy/utils/datatypes.py b/scrapy/utils/datatypes.py
index 807a95504..fd5ac3b08 100644
--- a/scrapy/utils/datatypes.py
+++ b/scrapy/utils/datatypes.py
@@ -94,8 +94,11 @@ class CaseInsensitiveDict(collections.UserDict):
 
     def __setitem__(self, key: AnyStr, value: Any) -> None:
         normalized_key = self._normkey(key)
-        if normalized_key.lower() in self._keys:
-            del self[self._keys[normalized_key.lower()]]
+        try:
+            lower_key = self._keys[normalized_key.lower()]
+            del self[lower_key]
+        except KeyError:
+            pass
         super().__setitem__(normalized_key, self._normvalue(value))
         self._keys[normalized_key.lower()] = normalized_key
 
diff --git a/tests/test_utils_datatypes.py b/tests/test_utils_datatypes.py
index 0a724f237..36df9006f 100644
--- a/tests/test_utils_datatypes.py
+++ b/tests/test_utils_datatypes.py
@@ -187,9 +187,15 @@ class CaseInsensitiveDictMixin:
     def test_copy(self):
         h1 = self.dict_class({'header1': 'value'})
         h2 = copy.copy(h1)
+        assert isinstance(h2, self.dict_class)
         self.assertEqual(h1, h2)
         self.assertEqual(h1.get('header1'), h2.get('header1'))
-        assert isinstance(h2, self.dict_class)
+        self.assertEqual(h1.get('header1'), h2.get('HEADER1'))
+        h3 = h1.copy()
+        assert isinstance(h3, self.dict_class)
+        self.assertEqual(h1, h3)
+        self.assertEqual(h1.get('header1'), h3.get('header1'))
+        self.assertEqual(h1.get('header1'), h3.get('HEADER1'))
 
 
 class CaseInsensitiveDictTest(CaseInsensitiveDictMixin, unittest.TestCase):