diff --git a/xarray/namedarray/core.py b/xarray/namedarray/core.py
index c535fcb0bc4..42950e1735c 100644
--- a/xarray/namedarray/core.py
+++ b/xarray/namedarray/core.py
@@ -835,6 +835,18 @@ def chunk(
                 for dim_number, dim in enumerate(self.dims)
                 if dim in chunks
             }
+            # Runs the same iteration as above, could extract to a shared helper
+            if any(val == "auto" for val in chunks.values()):
+                # If any chunks are "auto", we can add extra "auto" chunks for any
+                # zero-length dimensions to avoid dask errors. Need at least one existing
+                # auto or we might try to auto object dtype arrays incorrectly.
+                extra_autos = {
+                    dim_number: "auto"
+                    for dim_number, dim in enumerate(self.dims)
+                    if dim_number not in chunks  # keys are dim numbers now
+                    and self.sizes[dim] == 0
+                }
+                chunks = {**chunks, **extra_autos}
 
         chunkmanager = guess_chunkmanager(chunked_array_type)
 
diff --git a/xarray/tests/test_dask.py b/xarray/tests/test_dask.py
index 6b871aa79fb..3c0a4a19550 100644
--- a/xarray/tests/test_dask.py
+++ b/xarray/tests/test_dask.py
@@ -1210,6 +1210,18 @@ def test_auto_chunk_da(obj):
     assert actual.chunks == expected.chunks
 
 
+@pytest.mark.parametrize("obj", [make_da(), make_ds()])
+def test_partial_auto_chunk_zero_len_dim(obj):
+    obj = obj.isel(x=[])
+    obj.chunk({"y": "auto"})
+
+
+@pytest.mark.parametrize("obj", [make_ds()])
+def test_partial_auto_chunk_zero_len_dim_variable(obj):
+    obj = obj.isel(x=[])["a"].variable
+    obj.chunk({"y": "auto"})
+
+
 def test_auto_chunk_da_cftime():
     yrs = np.arange(2000, 2120)
     cftime_dates = xr.date_range(