diff --git a/sqlparse/filters/aligned_indent.py b/sqlparse/filters/aligned_indent.py
index 6ac99d62..abf369ef 100644
--- a/sqlparse/filters/aligned_indent.py
+++ b/sqlparse/filters/aligned_indent.py
@@ -80,9 +80,16 @@ def _process_case(self, tlist):
             # cond is None when 'else or end'
             stmt = cond[0] if cond else value[0]
 
+            # Malformed CASE expressions can yield case fragments that are no
+            # longer direct children of this token list. Aligned indentation is
+            # best-effort for such input, so do not try to insert relative to a
+            # token that token_index() cannot resolve.
+            if stmt not in tlist.tokens:
+                continue
+
             if i > 0:
                 tlist.insert_before(stmt, self.nl(offset_ - len(str(stmt))))
-            if cond:
+            if cond and cond[-1] in tlist.tokens:
                 ws = sql.Token(T.Whitespace, self.char * (
                     max_cond_width - condition_width[i]))
                 tlist.insert_after(cond[-1], ws)
diff --git a/tests/test_issue886.py b/tests/test_issue886.py
new file mode 100644
index 00000000..6c7503ee
--- /dev/null
+++ b/tests/test_issue886.py
@@ -0,0 +1,12 @@
+import sqlparse
+
+
+def test_reindent_aligned_malformed_case_does_not_crash():
+    sql = "CASE 'a' := WHERE END SELECT GO # ->>"
+
+    formatted = sqlparse.format(sql, reindent_aligned=True)
+
+    assert isinstance(formatted, str)
+    assert "CASE" in formatted
+    assert "END" in formatted
+    assert "SELECT" in formatted