diff --git a/python-package/xgboost/dask.py b/python-package/xgboost/dask.py
index 534ac0705..d05fe4951 100644
--- a/python-package/xgboost/dask.py
+++ b/python-package/xgboost/dask.py
@@ -802,6 +802,11 @@ async def _train_async(
     workers = list(_get_workers_from_data(dtrain, evals))
     _rabit_args = await _get_rabit_args(len(workers), client)
 
+    if params.get("booster", None) is not None and params["booster"] != "gbtree":
+        raise NotImplementedError(
+            f"booster `{params['booster']}` is not yet supported for dask."
+        )
+
     def dispatched_train(
         worker_addr: str,
         rabit_args: List[bytes],
diff --git a/tests/python/test_with_dask.py b/tests/python/test_with_dask.py
index e5481bf3b..5a66a5ec1 100644
--- a/tests/python/test_with_dask.py
+++ b/tests/python/test_with_dask.py
@@ -1167,6 +1167,19 @@ class TestWithDask:
             np.testing.assert_allclose(predt_0.compute(), predt_3)
 
 
+def test_unsupported_features(client: "Client"):
+    X, y, _ = generate_array()
+    # gblinear doesn't support distributed training.
+    with pytest.raises(NotImplementedError, match="gblinear"):
+        xgb.dask.train(
+            client, {"booster": "gblinear"}, xgb.dask.DaskDMatrix(client, X, y)
+        )
+    # dart prediction is not thread safe, running predict with each partition will have
+    # race.
+    with pytest.raises(NotImplementedError, match="dart"):
+        xgb.dask.train(client, {"booster": "dart"}, xgb.dask.DaskDMatrix(client, X, y))
+
+
 class TestDaskCallbacks:
     @pytest.mark.skipif(**tm.no_sklearn())
     def test_early_stopping(self, client: "Client") -> None: