diff --git a/doc/source/whatsnew/v2.1.0.rst b/doc/source/whatsnew/v2.1.0.rst
index 46f564f716b0a..cfce12c2930d7 100644
--- a/doc/source/whatsnew/v2.1.0.rst
+++ b/doc/source/whatsnew/v2.1.0.rst
@@ -350,8 +350,8 @@ Missing
 
 MultiIndex
 ^^^^^^^^^^
+- Bug in :meth:`DataFrame.__getitem__` not preserving dtypes for :class:`MultiIndex` partial keys (:issue:`51895`)
 - Bug in :meth:`MultiIndex.set_levels` not preserving dtypes for :class:`Categorical` (:issue:`52125`)
--
 
 I/O
 ^^^
diff --git a/pandas/core/frame.py b/pandas/core/frame.py
index 34e8c969d1b75..e04555ca30008 100644
--- a/pandas/core/frame.py
+++ b/pandas/core/frame.py
@@ -3831,18 +3831,8 @@ def _getitem_multilevel(self, key):
         if isinstance(loc, (slice, np.ndarray)):
             new_columns = self.columns[loc]
             result_columns = maybe_droplevels(new_columns, key)
-            if self._is_mixed_type:
-                result = self.reindex(columns=new_columns)
-                result.columns = result_columns
-            else:
-                new_values = self._values[:, loc]
-                result = self._constructor(
-                    new_values, index=self.index, columns=result_columns, copy=False
-                )
-                if using_copy_on_write() and isinstance(loc, slice):
-                    result._mgr.add_references(self._mgr)  # type: ignore[arg-type]
-
-                result = result.__finalize__(self)
+            result = self.iloc[:, loc]
+            result.columns = result_columns
 
             # If there is only one column being returned, and its name is
             # either an empty string, or a tuple with an empty string as its
diff --git a/pandas/tests/indexing/multiindex/test_multiindex.py b/pandas/tests/indexing/multiindex/test_multiindex.py
index d7285f0069c71..955c4acfd4c97 100644
--- a/pandas/tests/indexing/multiindex/test_multiindex.py
+++ b/pandas/tests/indexing/multiindex/test_multiindex.py
@@ -6,12 +6,14 @@
 
 import pandas as pd
 from pandas import (
+    CategoricalDtype,
     DataFrame,
     Index,
     MultiIndex,
     Series,
 )
 import pandas._testing as tm
+from pandas.core.arrays.boolean import BooleanDtype
 
 
 class TestMultiIndexBasic:
@@ -207,6 +209,24 @@ def test_multiindex_with_na_missing_key(self):
         with pytest.raises(KeyError, match="missing_key"):
             df[[("missing_key",)]]
 
+    def test_multiindex_dtype_preservation(self):
+        # GH51261
+        columns = MultiIndex.from_tuples([("A", "B")], names=["lvl1", "lvl2"])
+        df = DataFrame(["value"], columns=columns).astype("category")
+        df_no_multiindex = df["A"]
+        assert isinstance(df_no_multiindex["B"].dtype, CategoricalDtype)
+
+        # geopandas 1763 analogue
+        df = DataFrame(
+            [[1, 0], [0, 1]],
+            columns=[
+                ["foo", "foo"],
+                ["location", "location"],
+                ["x", "y"],
+            ],
+        ).assign(bools=Series([True, False], dtype="boolean"))
+        assert isinstance(df["bools"].dtype, BooleanDtype)
+
     def test_multiindex_from_tuples_with_nan(self):
         # GH#23578
         result = MultiIndex.from_tuples([("a", "b", "c"), np.nan, ("d", "", "")])