diff --git a/doc/source/whatsnew/v0.24.2.rst b/doc/source/whatsnew/v0.24.2.rst
index a7e522d27f8e26..8f4beb3f484a47 100644
--- a/doc/source/whatsnew/v0.24.2.rst
+++ b/doc/source/whatsnew/v0.24.2.rst
@@ -96,7 +96,7 @@ Bug Fixes
 **Other**
 
 - Bug in :meth:`Series.is_unique` where single occurrences of ``NaN`` were not considered unique (:issue:`25180`)
--
+- Bug in :func:`merge` when merging an empty ``DataFrame`` with an ``Int64`` column or a non-empty ``DataFrame`` with an ``Int64`` column that is all ``NaN`` (:issue:`25183`)
 -
 
 .. _whatsnew_0.242.contributors:
diff --git a/pandas/core/internals/concat.py b/pandas/core/internals/concat.py
index 640587b7f9f314..cb982749626564 100644
--- a/pandas/core/internals/concat.py
+++ b/pandas/core/internals/concat.py
@@ -190,6 +190,8 @@ def get_reindexed_values(self, empty_dtype, upcasted_na):
                     pass
                 elif getattr(self.block, 'is_sparse', False):
                     pass
+                elif getattr(self.block, 'is_extension', False):
+                    pass
                 else:
                     missing_arr = np.empty(self.shape, dtype=empty_dtype)
                     missing_arr.fill(fill_value)
diff --git a/pandas/tests/reshape/merge/test_merge.py b/pandas/tests/reshape/merge/test_merge.py
index 25487ccc76e624..7a97368504fd61 100644
--- a/pandas/tests/reshape/merge/test_merge.py
+++ b/pandas/tests/reshape/merge/test_merge.py
@@ -39,6 +39,54 @@ def get_test_data(ngroups=NGROUPS, n=N):
     return arr
 
 
+def get_series():
+    return [
+        pd.Series([1], dtype='int64'),
+        pd.Series([1], dtype='Int64'),
+        pd.Series([1.23]),
+        pd.Series(['foo']),
+        pd.Series([True]),
+        pd.Series([pd.Timestamp('2018-01-01')]),
+        pd.Series([pd.Timestamp('2018-01-01', tz='US/Eastern')]),
+    ]
+
+
+def get_series_na():
+    return [
+        pd.Series([np.nan], dtype='Int64'),
+        pd.Series([np.nan], dtype='float'),
+        pd.Series([np.nan], dtype='object'),
+        pd.Series([pd.NaT]),
+    ]
+
+
+@pytest.fixture(params=get_series(), ids=lambda x: x.dtype.name)
+def series_of_dtype(request):
+    """
+    A parametrized fixture returning a variety of Series of different
+    dtypes
+    """
+    return request.param
+
+
+@pytest.fixture(params=get_series(), ids=lambda x: x.dtype.name)
+def series_of_dtype2(request):
+    """
+    A duplicate of the series_of_dtype fixture, so that it can be used
+    twice by a single function
+    """
+    return request.param
+
+
+@pytest.fixture(params=get_series_na(), ids=lambda x: x.dtype.name)
+def series_of_dtype_all_na(request):
+    """
+    A parametrized fixture returning a variety of Series with all NA
+    values
+    """
+    return request.param
+
+
 class TestMerge(object):
 
     def setup_method(self, method):
@@ -428,6 +476,36 @@ def check2(exp, kwarg):
                 check1(exp_in, kwarg)
                 check2(exp_out, kwarg)
 
+    def test_merge_empty_frame(self, series_of_dtype, series_of_dtype2):
+        # GH 25183
+        df = pd.DataFrame({'key': series_of_dtype, 'value': series_of_dtype2},
+                          columns=['key', 'value'])
+        df_empty = df[:0]
+        expected = pd.DataFrame({
+            'value_x': pd.Series(dtype=df.dtypes['value']),
+            'key': pd.Series(dtype=df.dtypes['key']),
+            'value_y': pd.Series(dtype=df.dtypes['value']),
+        }, columns=['value_x', 'key', 'value_y'])
+        actual = df_empty.merge(df, on='key')
+        assert_frame_equal(actual, expected)
+
+    def test_merge_all_na_column(self, series_of_dtype,
+                                 series_of_dtype_all_na):
+        # GH 25183
+        df_left = pd.DataFrame(
+            {'key': series_of_dtype, 'value': series_of_dtype_all_na},
+            columns=['key', 'value'])
+        df_right = pd.DataFrame(
+            {'key': series_of_dtype, 'value': series_of_dtype_all_na},
+            columns=['key', 'value'])
+        expected = pd.DataFrame({
+            'key': series_of_dtype,
+            'value_x': series_of_dtype_all_na,
+            'value_y': series_of_dtype_all_na,
+        }, columns=['key', 'value_x', 'value_y'])
+        actual = df_left.merge(df_right, on='key')
+        assert_frame_equal(actual, expected)
+
     def test_merge_nosort(self):
         # #2098, anything to do?