diff --git a/doc/source/whatsnew/v0.15.1.txt b/doc/source/whatsnew/v0.15.1.txt
index 04aec92f448c7..af312e3e22275 100644
--- a/doc/source/whatsnew/v0.15.1.txt
+++ b/doc/source/whatsnew/v0.15.1.txt
@@ -118,7 +118,8 @@ Bug Fixes
 
 
 
-- Bug in numeric index operations of add/sub with Float/Index Index with numpy arrays (:issue:`8608`)
+- Bug in numeric index operations of add/sub with Float/Index Index with numpy arrays (:issue:`8608`
+- Bug in setitem with empty indexer and unwanted coercion of dtypes (:issue:`8669`)
 
 
 
diff --git a/pandas/core/internals.py b/pandas/core/internals.py
index 89e1cd6ce0fb6..6dd123996b125 100644
--- a/pandas/core/internals.py
+++ b/pandas/core/internals.py
@@ -549,10 +549,31 @@ def setitem(self, indexer, value):
                                      "different length than the value")
 
         try:
+
+            def _is_scalar_indexer(indexer):
+                # treat a len 0 array like a scalar
+                # return True if we are all scalar indexers
+
+                if arr_value.ndim == 1:
+                    if not isinstance(indexer, tuple):
+                        indexer = tuple([indexer])
+
+                    def _is_ok(idx):
+
+                        if np.isscalar(idx):
+                            return True
+                        elif isinstance(idx, slice):
+                            return False
+                        return len(idx) == 0
+
+                    return all([ _is_ok(idx) for idx in indexer ])
+                return False
+
+
             # setting a single element for each dim and with a rhs that could be say a list
-            # GH 6043
-            if arr_value.ndim == 1 and (
-                np.isscalar(indexer) or (isinstance(indexer, tuple) and all([ np.isscalar(idx) for idx in indexer ]))):
+            # or empty indexers (so no astyping)
+            # GH 6043, 8669 (empty)
+            if _is_scalar_indexer(indexer):
                 values[indexer] = value
 
             # if we are an exact match (ex-broadcasting),
diff --git a/pandas/tests/test_indexing.py b/pandas/tests/test_indexing.py
index a2f3284efcb82..7bc5ca0bfb2e7 100644
--- a/pandas/tests/test_indexing.py
+++ b/pandas/tests/test_indexing.py
@@ -1043,6 +1043,13 @@ def test_loc_setitem_frame(self):
         expected = DataFrame(dict(A = Series(val1,index=keys1), B = Series(val2,index=keys2))).reindex(index=index)
         assert_frame_equal(df, expected)
 
+        # GH 8669
+        # invalid coercion of nan -> int
+        df = DataFrame({'A' : [1,2,3], 'B' : np.nan })
+        df.loc[df.B > df.A, 'B'] = df.A
+        expected = DataFrame({'A' : [1,2,3], 'B' : np.nan})
+        assert_frame_equal(df, expected)
+
         # GH 6546
         # setting with mixed labels
         df = DataFrame({1:[1,2],2:[3,4],'a':['a','b']})
@@ -1055,7 +1062,6 @@ def test_loc_setitem_frame(self):
         df.loc[0,[1,2]] = [5,6]
         assert_frame_equal(df, expected)
 
-
     def test_loc_setitem_frame_multiples(self):
         # multiple setting
         df = DataFrame({ 'A' : ['foo','bar','baz'],