diff --git a/doc/source/whatsnew/v2.0.0.rst b/doc/source/whatsnew/v2.0.0.rst
index 0ceda331de790..0ea8a5db4afb3 100644
--- a/doc/source/whatsnew/v2.0.0.rst
+++ b/doc/source/whatsnew/v2.0.0.rst
@@ -600,6 +600,7 @@ Other API changes
   methods to get a full slice (for example ``df.loc[:]`` or ``df[:]``) (:issue:`49469`)
 - Disallow computing ``cumprod`` for :class:`Timedelta` object; previously this returned incorrect values (:issue:`50246`)
 - Loading a JSON file with duplicate columns using ``read_json(orient='split')`` renames columns to avoid duplicates, as :func:`read_csv` and the other readers do (:issue:`50370`)
+- The levels of the index of the :class:`Series` returned from ``Series.sparse.from_coo`` now always have dtype ``int32``. Previously they had dtype ``int64`` (:issue:`50926`)
 - :func:`to_datetime` with ``unit`` of either "Y" or "M" will now raise if a sequence contains a non-round ``float`` value, matching the ``Timestamp`` behavior (:issue:`50301`)
 -
 
diff --git a/pandas/core/arrays/sparse/scipy_sparse.py b/pandas/core/arrays/sparse/scipy_sparse.py
index 3ef5ba5835d79..8e3ba57cbaba4 100644
--- a/pandas/core/arrays/sparse/scipy_sparse.py
+++ b/pandas/core/arrays/sparse/scipy_sparse.py
@@ -203,9 +203,6 @@ def coo_to_sparse_series(
     ser = ser.sort_index()
     ser = ser.astype(SparseDtype(ser.dtype))
     if dense_index:
-        # is there a better constructor method to use here?
-        i = range(A.shape[0])
-        j = range(A.shape[1])
-        ind = MultiIndex.from_product([i, j])
+        ind = MultiIndex.from_product([A.row, A.col])
         ser = ser.reindex(ind)
     return ser
diff --git a/pandas/tests/arrays/sparse/test_accessor.py b/pandas/tests/arrays/sparse/test_accessor.py
index 9ac0d9d0401ed..7d6a9e18a26c6 100644
--- a/pandas/tests/arrays/sparse/test_accessor.py
+++ b/pandas/tests/arrays/sparse/test_accessor.py
@@ -218,14 +218,11 @@ def test_series_from_coo(self, dtype, dense_index):
         A = scipy.sparse.eye(3, format="coo", dtype=dtype)
         result = pd.Series.sparse.from_coo(A, dense_index=dense_index)
 
-        # TODO: GH49560: scipy.sparse.eye always has A.row and A.col dtype as int32.
-        # fix index_dtype to follow scipy.sparse convention (always int32)?
-        index_dtype = np.int64 if dense_index else np.int32
         index = pd.MultiIndex.from_tuples(
             [
-                np.array([0, 0], dtype=index_dtype),
-                np.array([1, 1], dtype=index_dtype),
-                np.array([2, 2], dtype=index_dtype),
+                np.array([0, 0], dtype=np.int32),
+                np.array([1, 1], dtype=np.int32),
+                np.array([2, 2], dtype=np.int32),
             ],
         )
         expected = pd.Series(SparseArray(np.array([1, 1, 1], dtype=dtype)), index=index)