diff --git a/packages/zarr-metadata/changes/422.bugfix.md b/packages/zarr-metadata/changes/422.bugfix.md
new file mode 100644
index 0000000000..95db2dbe70
--- /dev/null
+++ b/packages/zarr-metadata/changes/422.bugfix.md
@@ -0,0 +1 @@
+The Zarr format 2 time dtypes accept the generic time unit in brackets — `<M8[generic]`, or `<m8[2generic]` with a scale — the spellings zarr-python writes for a NumPy generic-unit time dtype. A unit is still required: the unitless `<M8` stays refused, as the v2 spec demands.
diff --git a/packages/zarr-metadata/src/zarr_metadata/_numpy.py b/packages/zarr-metadata/src/zarr_metadata/_numpy.py
index 84a6705c7f..97c02e2c5e 100644
--- a/packages/zarr-metadata/src/zarr_metadata/_numpy.py
+++ b/packages/zarr-metadata/src/zarr_metadata/_numpy.py
@@ -24,8 +24,9 @@
     "ps",
     "fs",
     "as",
+    "generic",
 )
-"""The units NumPy accepts for a datetime64 or timedelta64 (https://numpy.org/doc/stable/reference/arrays.datetime.html#datetime-units). The v3 types also accept `"generic"`."""
+"""The units NumPy accepts for a datetime64 or timedelta64 (https://numpy.org/doc/stable/reference/arrays.datetime.html#datetime-units), and `"generic"`, NumPy's unit-less unit, which NumPy parses in brackets (`<M8[generic]`) though its own `dtype.str` omits it. The v2 spec's unit table does not list `generic`, but zarr-python writes it (zarr-developers/zarr-python#4580)."""
 
 NUMPY_TIME_MAX_SCALE_FACTOR: Final = 2**31 - 1
 """The largest `scale_factor` that numpy stores, because the field is a signed int32."""
diff --git a/packages/zarr-metadata/src/zarr_metadata/v3/data_type/_numpy_time.py b/packages/zarr-metadata/src/zarr_metadata/v3/data_type/_numpy_time.py
index e8568863f4..ed631a353f 100644
--- a/packages/zarr-metadata/src/zarr_metadata/v3/data_type/_numpy_time.py
+++ b/packages/zarr-metadata/src/zarr_metadata/v3/data_type/_numpy_time.py
@@ -14,7 +14,7 @@
 ]
 """The time unit codes of the numpy time types, `numpy.datetime64` and `numpy.timedelta64`."""
 
-NUMPY_TIME_UNIT: Final = (*NUMPY_TIME_UNITS, "generic")
+NUMPY_TIME_UNIT: Final = NUMPY_TIME_UNITS
 """Runtime tuple of the permitted `numpy.timedelta64`/`numpy.datetime64` unit strings."""
 
 numpy_time_storage: Final = multi_byte
diff --git a/packages/zarr-metadata/tests/v2/test_data_types.py b/packages/zarr-metadata/tests/v2/test_data_types.py
index ea0569ffaa..1438889a2e 100644
--- a/packages/zarr-metadata/tests/v2/test_data_types.py
+++ b/packages/zarr-metadata/tests/v2/test_data_types.py
@@ -59,7 +59,9 @@ def _read(value: object) -> tuple[type, list[tuple[Loc, str]]]:
         ("<M8[ns]", "datetime64", "<M8[ns]"),
         (">M8[10s]", "datetime64", ">M8[10s]"),
         ("<M8[μs]", "datetime64", "<M8[us]"),
+        ("<M8[generic]", "datetime64", "<M8[generic]"),
         ("<m8[D]", "timedelta64", "<m8[D]"),
+        (">m8[2generic]", "timedelta64", ">m8[2generic]"),
         ([["x", "<f4"], ["y", "<i4", [2]]], "struct", (("x", "<f4"), ("y", "<i4", (2,)))),
         ([["r", "|u1"], ["s", [["t", "<f4"]]]], "struct", (("r", "|u1"), ("s", (("t", "<f4"),)))),
         # NumPy names the padding fields of an aligned struct "", and zarr 2.x writes them.