diff --git a/docs/release.rst b/docs/release.rst
index 12e1449a80..a361bbc8d9 100644
--- a/docs/release.rst
+++ b/docs/release.rst
@@ -21,11 +21,14 @@ Unreleased
 Maintenance
 ~~~~~~~~~~~
 
+* Preserve ``dimension_separator`` when resizing arrays.
+  By :user:`Ziwen Liu <ziw-liu>` :issue:`1533`.
+
 * Initialise some sets in tests with set literals instead of list literals.
   By :user:`Dimitri Papadopoulos Orfanos <DimitriPapadopoulos>` :issue:`1534`.
 
 * Allow ``black`` code formatter to be run with any Python version.
-  By :user:`David Stansby <dstansby>` :issue:`1549`
+  By :user:`David Stansby <dstansby>` :issue:`1549`.
 
 .. _release_2.16.1:
 
diff --git a/zarr/core.py b/zarr/core.py
index 43ccdbaf7d..2177e9055c 100644
--- a/zarr/core.py
+++ b/zarr/core.py
@@ -253,7 +253,6 @@ def _load_metadata_nosync(self):
         except KeyError:
             raise ArrayNotFoundError(self._path)
         else:
-
             # decode and store metadata as instance members
             meta = self._store._metadata_class.decode_array_metadata(meta_bytes)
             self._meta = meta
@@ -341,7 +340,14 @@ def _flush_metadata_nosync(self):
             filters=filters_config,
         )
         if getattr(self._store, "_store_version", 2) == 2:
-            meta.update(dict(chunks=self._chunks, dtype=self._dtype, order=self._order))
+            meta.update(
+                dict(
+                    chunks=self._chunks,
+                    dtype=self._dtype,
+                    order=self._order,
+                    dimension_separator=self._dimension_separator,
+                )
+            )
         else:
             meta.update(
                 dict(
@@ -1358,7 +1364,6 @@ def get_mask_selection(self, selection, out=None, fields=None):
         return self._get_selection(indexer=indexer, out=out, fields=fields)
 
     def _get_selection(self, indexer, out=None, fields=None):
-
         # We iterate over all chunks which overlap the selection and thus contain data
         # that needs to be extracted. Each chunk is processed in turn, extracting the
         # necessary data and storing into the correct location in the output array.
@@ -1983,7 +1988,6 @@ def _set_basic_selection_nd(self, selection, value, fields=None):
         self._set_selection(indexer, value, fields=fields)
 
     def _set_selection(self, indexer, value, fields=None):
-
         # We iterate over all chunks which overlap the selection and thus contain data
         # that needs to be replaced. Each chunk is processed in turn, extracting the
         # necessary data from the value array and storing into the chunk array.
@@ -2018,7 +2022,6 @@ def _set_selection(self, indexer, value, fields=None):
         ):
             # iterative approach
             for chunk_coords, chunk_selection, out_selection in indexer:
-
                 # extract data to store
                 if sel_shape == ():
                     chunk_value = value
@@ -2077,7 +2080,6 @@ def _process_chunk(
             and not self._filters
             and self._dtype != object
         ):
-
             dest = out[out_selection]
             # Assume that array-like objects that doesn't have a
             # `writeable` flag is writable.
@@ -2088,7 +2090,6 @@ def _process_chunk(
             )
 
             if write_direct:
-
                 # optimization: we want the whole chunk, and the destination is
                 # contiguous, so we can decompress directly from the chunk
                 # into the destination array
@@ -2321,7 +2322,6 @@ def _process_for_setitem(self, ckey, chunk_selection, value, fields=None):
             # to access the existing chunk data
 
             if is_scalar(value, self._dtype):
-
                 # setup array filled with value
                 chunk = np.empty_like(
                     self._meta_array, shape=self._chunks, dtype=self._dtype, order=self._order
@@ -2329,7 +2329,6 @@ def _process_for_setitem(self, ckey, chunk_selection, value, fields=None):
                 chunk.fill(value)
 
             else:
-
                 # ensure array is contiguous
                 chunk = value.astype(self._dtype, order=self._order, copy=False)
 
@@ -2337,12 +2336,10 @@ def _process_for_setitem(self, ckey, chunk_selection, value, fields=None):
             # partially replace the contents of this chunk
 
             try:
-
                 # obtain compressed data for chunk
                 cdata = self.chunk_store[ckey]
 
             except KeyError:
-
                 # chunk not initialized
                 if self._fill_value is not None:
                     chunk = np.empty_like(
@@ -2359,7 +2356,6 @@ def _process_for_setitem(self, ckey, chunk_selection, value, fields=None):
                     )
 
             else:
-
                 # decode chunk
                 chunk = self._decode_chunk(cdata)
                 if not chunk.flags.writeable:
@@ -2429,7 +2425,6 @@ def _decode_chunk(self, cdata, start=None, nitems=None, expected_shape=None):
         return chunk
 
     def _encode_chunk(self, chunk):
-
         # apply filters
         if self._filters:
             for f in self._filters:
@@ -2619,7 +2614,6 @@ def __setstate__(self, state):
         self.__init__(**state)
 
     def _synchronized_op(self, f, *args, **kwargs):
-
         if self._synchronizer is None:
             # no synchronization
             lock = nolock
@@ -2636,7 +2630,6 @@ def _synchronized_op(self, f, *args, **kwargs):
         return result
 
     def _write_op(self, f, *args, **kwargs):
-
         # guard condition
         if self._read_only:
             raise ReadOnlyError()
@@ -2676,7 +2669,6 @@ def resize(self, *args):
         return self._write_op(self._resize_nosync, *args)
 
     def _resize_nosync(self, *args):
-
         # normalize new shape argument
         old_shape = self._shape
         new_shape = normalize_resize_args(old_shape, *args)
@@ -2755,7 +2747,6 @@ def append(self, data, axis=0):
         return self._write_op(self._append_nosync, data, axis=axis)
 
     def _append_nosync(self, data, axis=0):
-
         # ensure data is array-like
         if not hasattr(data, "shape"):
             data = np.asanyarray(data, like=self._meta_array)
diff --git a/zarr/tests/test_core.py b/zarr/tests/test_core.py
index 77b9415d8b..f3ca73dea8 100644
--- a/zarr/tests/test_core.py
+++ b/zarr/tests/test_core.py
@@ -132,7 +132,6 @@ def create_array(self, shape: Union[int, Tuple[int, ...]], **kwargs):
         return Array(store, **access_array_kwargs)
 
     def test_array_init(self):
-
         # normal initialization
         store = self.create_store()
         init_array(store, shape=100, chunks=10, dtype="<f8")
@@ -328,7 +327,6 @@ def test_array_1d(self):
 
     def test_array_1d_fill_value(self):
         for fill_value in -1, 0, 1, 10:
-
             a = np.arange(1050)
             f = np.empty_like(a)
             f.fill(fill_value)
@@ -575,7 +573,6 @@ def test_array_2d_partial(self):
         z.store.close()
 
     def test_array_order(self):
-
         # 1D
         a = np.arange(1050)
         for order in "C", "F":
@@ -667,7 +664,6 @@ def test_hexdigest(self):
         assert self.expected() == found
 
     def test_resize_1d(self):
-
         z = self.create_array(shape=105, chunks=10, dtype="i4", fill_value=0)
         a = np.arange(105, dtype="i4")
         z[:] = a
@@ -703,7 +699,6 @@ def test_resize_1d(self):
         z.store.close()
 
     def test_resize_2d(self):
-
         z = self.create_array(shape=(105, 105), chunks=(10, 10), dtype="i4", fill_value=0)
         a = np.arange(105 * 105, dtype="i4").reshape((105, 105))
         z[:] = a
@@ -756,8 +751,14 @@ def test_resize_2d(self):
 
         z.store.close()
 
-    def test_append_1d(self):
+        # checks that resizing preserves metadata
+        if self.dimension_separator == "/":
+            z_ = zarr.open(z.store)
+            if hasattr(z_, "dimension_separator"):
+                assert z_.dimension_separator == self.dimension_separator
+            z_.store.close()
 
+    def test_append_1d(self):
         a = np.arange(105)
         z = self.create_array(shape=a.shape, chunks=10, dtype=a.dtype)
         z[:] = a
@@ -786,7 +787,6 @@ def test_append_1d(self):
         z.store.close()
 
     def test_append_2d(self):
-
         a = np.arange(105 * 105, dtype="i4").reshape((105, 105))
         z = self.create_array(shape=a.shape, chunks=(10, 10), dtype=a.dtype)
         z[:] = a
@@ -808,7 +808,6 @@ def test_append_2d(self):
         z.store.close()
 
     def test_append_2d_axis(self):
-
         a = np.arange(105 * 105, dtype="i4").reshape((105, 105))
         z = self.create_array(shape=a.shape, chunks=(10, 10), dtype=a.dtype)
         z[:] = a
@@ -837,7 +836,6 @@ def test_append_bad_shape(self):
         z.store.close()
 
     def test_read_only(self):
-
         z = self.create_array(shape=1000, chunks=100)
         assert not z.read_only
         z.store.close()
@@ -868,7 +866,6 @@ def test_read_only(self):
         z.store.close()
 
     def test_pickle(self):
-
         # setup array
         z = self.create_array(
             shape=1000, chunks=100, dtype=int, cache_metadata=False, cache_attrs=False
@@ -1082,7 +1079,6 @@ def test_nchunks_initialized(self):
             z.store.close()
 
     def test_array_dtype_shape(self):
-
         dt = "(2, 2)f4"
         # setup some data
         d = np.array([((0, 1), (1, 2)), ((1, 2), (2, 3)), ((2, 3), (3, 4))], dtype=dt)
@@ -1175,7 +1171,6 @@ def test_structured_array_nested(self):
         self.check_structured_array(d, fill_values)
 
     def test_dtypes(self):
-
         # integers
         for dtype in "u1", "u2", "u4", "u8", "i1", "i2", "i4", "i8":
             z = self.create_array(shape=10, chunks=3, dtype=dtype)
@@ -1237,7 +1232,6 @@ def test_dtypes(self):
             self.create_array(shape=100, dtype="m8")
 
     def test_object_arrays(self):
-
         # an object_codec is required for object arrays
         with pytest.raises(ValueError):
             self.create_array(shape=10, chunks=3, dtype=object)
@@ -1297,7 +1291,6 @@ def test_object_arrays(self):
         z.store.close()
 
     def test_object_arrays_vlen_text(self):
-
         data = np.array(greetings * 1000, dtype=object)
         z = self.create_array(shape=data.shape, dtype=object, object_codec=VLenUTF8())
         z[0] = "foo"
@@ -1343,7 +1336,6 @@ def test_object_arrays_vlen_text(self):
         z.store.close()
 
     def test_object_arrays_vlen_bytes(self):
-
         greetings_bytes = [g.encode("utf8") for g in greetings]
         data = np.array(greetings_bytes * 1000, dtype=object)
 
@@ -1374,7 +1366,6 @@ def test_object_arrays_vlen_bytes(self):
         z.store.close()
 
     def test_object_arrays_vlen_array(self):
-
         data = np.array(
             [np.array([1, 3, 7]), np.array([5]), np.array([2, 8, 12])] * 1000, dtype=object
         )
@@ -1410,7 +1401,6 @@ def compare_arrays(expected, actual, item_dtype):
             z.store.close()
 
     def test_object_arrays_danger(self):
-
         # do something dangerous - manually force an object array with no object codec
         z = self.create_array(shape=5, chunks=2, dtype=object, fill_value=0, object_codec=MsgPack())
         z._filters = None  # wipe filters
@@ -1438,7 +1428,6 @@ def test_object_arrays_danger(self):
             z.store.close()
 
     def test_object_codec_warnings(self):
-
         with pytest.warns(UserWarning):
             # provide object_codec, but not object dtype
             z = self.create_array(shape=10, chunks=5, dtype="i4", object_codec=JSON())
@@ -1449,7 +1438,6 @@ def test_object_codec_warnings(self):
         "unsupported numpy version",
     )
     def test_structured_array_contain_object(self):
-
         if "PartialRead" in self.__class__.__name__:
             pytest.skip("partial reads of object arrays not supported")
 
@@ -1631,7 +1619,6 @@ def expected(self):
         ]
 
     def test_nbytes_stored(self):
-
         # MemoryStore as store
         z = self.create_array(shape=1000, chunks=100)
         expect_nbytes_stored = sum(
@@ -1665,7 +1652,6 @@ def expected(self):
         ]
 
     def test_nbytes_stored(self):
-
         z = self.create_array(shape=1000, chunks=100)
         expect_nbytes_stored = sum(buffer_size(v) for v in z.store.values())
         expect_nbytes_stored += sum(buffer_size(v) for v in z.chunk_store.values())
@@ -1688,7 +1674,6 @@ def create_store(self):
         return store
 
     def test_nbytes_stored(self):
-
         # dict as store
         z = self.create_array(shape=1000, chunks=100)
         expect_nbytes_stored = sum(buffer_size(v) for v in z.store.values())
@@ -1844,7 +1829,6 @@ def test_nchunks_initialized(self):
         assert 0 == z.nchunks_initialized
 
     def test_array_order(self):
-
         # N5 only supports 'C' at the moment
         with pytest.raises(ValueError):
             self.create_array(shape=(10, 11), chunks=(10, 11), dtype="i8", order="F")
@@ -1907,7 +1891,6 @@ def test_structured_array_nested(self):
             self.check_structured_array(d, fill_values)
 
     def test_dtypes(self):
-
         # integers
         for dtype in "u1", "u2", "u4", "u8", "i1", "i2", "i4", "i8":
             z = self.create_array(shape=10, chunks=3, dtype=dtype)
@@ -1931,7 +1914,6 @@ def test_dtypes(self):
             self.create_array(shape=100, dtype="m8")
 
     def test_object_arrays(self):
-
         # an object_codec is required for object arrays
         with pytest.raises(ValueError):
             self.create_array(shape=10, chunks=3, dtype=object)
@@ -1947,7 +1929,6 @@ def test_object_arrays(self):
             self.create_array(shape=10, chunks=3, dtype=object, object_codec=MsgPack())
 
     def test_object_arrays_vlen_text(self):
-
         data = np.array(greetings * 1000, dtype=object)
 
         with pytest.raises(ValueError):
@@ -1958,7 +1939,6 @@ def test_object_arrays_vlen_text(self):
             self.create_array(shape=data.shape, dtype=str)
 
     def test_object_arrays_vlen_bytes(self):
-
         greetings_bytes = [g.encode("utf8") for g in greetings]
         data = np.array(greetings_bytes * 1000, dtype=object)
 
@@ -1970,7 +1950,6 @@ def test_object_arrays_vlen_bytes(self):
             self.create_array(shape=data.shape, dtype=bytes)
 
     def test_object_arrays_vlen_array(self):
-
         data = np.array(
             [np.array([1, 3, 7]), np.array([5]), np.array([2, 8, 12])] * 1000, dtype=object
         )
@@ -2122,7 +2101,6 @@ def expected(self):
 
 
 class TestArrayWithBZ2Compressor(TestArray):
-
     compressor = BZ2(level=1)
 
     def expected(self):
@@ -2136,7 +2114,6 @@ def expected(self):
 
 
 class TestArrayWithBloscCompressor(TestArray):
-
     compressor = Blosc(cname="zstd", clevel=1, shuffle=1)
 
     def expected(self):
@@ -2157,7 +2134,6 @@ def expected(self):
 
 @unittest.skipIf(LZMA is None, "LZMA codec not available")
 class TestArrayWithLZMACompressor(TestArray):
-
     compressor = LZMA(preset=1)
 
     def expected(self):
@@ -2171,7 +2147,6 @@ def expected(self):
 
 
 class TestArrayWithFilters(TestArray):
-
     compressor = Zlib(1)
 
     def create_filters(self, dtype: Optional[str]) -> Tuple[Any, ...]:
@@ -2583,7 +2558,6 @@ def expected(self):
 @pytest.mark.skipif(not v3_api_available, reason="V3 is disabled")
 class TestArrayWithPathV3(TestArrayV3):
     def test_array_init(self):
-
         store = self.create_store()
         # can initialize an array without a path
         init_array(store, shape=100, chunks=10, dtype="<f8")
@@ -2646,7 +2620,6 @@ def expected(self):
         ]
 
     def test_nbytes_stored(self):
-
         # dict as store
         z = self.create_array(shape=1000, chunks=100)
         expect_nbytes_stored = sum(buffer_size(v) for k, v in z.store.items() if k != "zarr.json")
@@ -2664,7 +2637,6 @@ def test_nbytes_stored(self):
         z.store.close()
 
     def test_view(self):
-
         # dict as store
         z = self.create_array(shape=1005, chunks=100, dtype=float)
 
@@ -2722,7 +2694,6 @@ def expected(self):
         ]
 
     def test_nbytes_stored(self):
-
         z = self.create_array(shape=1000, chunks=100)
         expect_nbytes_stored = sum(buffer_size(v) for k, v in z.store.items() if k != "zarr.json")
         expect_nbytes_stored += sum(
@@ -2901,7 +2872,6 @@ def test_nbytes_stored(self):
         assert expect_nbytes_stored == z.nbytes_stored
 
     def test_len(self):
-
         # dict as store
         z = self.create_array(shape=1000, chunks=100)
         assert len(z._store) == 2