diff --git a/ibis/backends/sql/rewrites.py b/ibis/backends/sql/rewrites.py
index 783ff10ea968..554508a3d8ff 100644
--- a/ibis/backends/sql/rewrites.py
+++ b/ibis/backends/sql/rewrites.py
@@ -102,6 +102,12 @@ def dtype(self):
         return self.func.dtype
 
 
+# TODO(kszucs): there is a better strategy to rewrite the relational operations
+# to Select nodes by wrapping the leaf nodes in a Select node and then merging
+# Project, Filter, Sort, etc. incrementally into the Select node. This way we
+# can have tighter control over simplification logic.
+
+
 @replace(p.Project)
 def project_to_select(_, **kwargs):
     """Convert a Project node to a Select node."""
@@ -164,20 +170,20 @@ def merge_select_select(_, **kwargs):
     for v in _.predicates:
         if v.find((ops.ExistsSubquery, ops.InSubquery), filter=ops.Value):
             return _
+    # cannot merge if the inner select has different order keys than the outer
+    if _.sort_keys and _.parent.sort_keys:
+        return _
 
     subs = {ops.Field(_.parent, k): v for k, v in _.parent.values.items()}
     selections = {k: v.replace(subs, filter=ops.Value) for k, v in _.selections.items()}
     predicates = tuple(p.replace(subs, filter=ops.Value) for p in _.predicates)
     sort_keys = tuple(s.replace(subs, filter=ops.Value) for s in _.sort_keys)
 
-    unique_predicates = toolz.unique(_.parent.predicates + predicates)
-    unique_sort_keys = {s.expr: s for s in _.parent.sort_keys + sort_keys}
-
     return Select(
         _.parent.parent,
         selections=selections,
-        predicates=unique_predicates,
-        sort_keys=unique_sort_keys.values(),
+        predicates=toolz.unique(_.parent.predicates + predicates),
+        sort_keys=_.parent.sort_keys + sort_keys,
     )
 
 
diff --git a/ibis/backends/tests/sql/snapshots/test_sql/test_double_order_by_not_fused/out.sql b/ibis/backends/tests/sql/snapshots/test_sql/test_double_order_by_not_fused/out.sql
new file mode 100644
index 000000000000..38ceb36268fc
--- /dev/null
+++ b/ibis/backends/tests/sql/snapshots/test_sql/test_double_order_by_not_fused/out.sql
@@ -0,0 +1,13 @@
+SELECT
+  "t1"."a",
+  "t1"."b"
+FROM (
+  SELECT
+    "t0"."a",
+    "t0"."b"
+  FROM "t" AS "t0"
+  ORDER BY
+    "t0"."a" ASC
+) AS "t1"
+ORDER BY
+  "t1"."b" DESC
\ No newline at end of file
diff --git a/ibis/backends/tests/sql/test_sql.py b/ibis/backends/tests/sql/test_sql.py
index d4bb8fe8033e..b52f97451f9e 100644
--- a/ibis/backends/tests/sql/test_sql.py
+++ b/ibis/backends/tests/sql/test_sql.py
@@ -518,6 +518,12 @@ def test_order_by_expr(snapshot):
     snapshot.assert_match(to_sql(expr), "out.sql")
 
 
+def test_double_order_by_not_fused(snapshot):
+    t = ibis.table(dict(a="int", b="string"), name="t")
+    expr = t.order_by(t.a).order_by(t.b.desc())
+    snapshot.assert_match(to_sql(expr), "out.sql")
+
+
 def test_no_cartesian_join(snapshot):
     customers = ibis.table(
         dict(customer_id="int64", first_name="string", last_name="string"),