diff --git a/CHANGELOG b/CHANGELOG
index 44d5938e..4dbaada1 100644
--- a/CHANGELOG
+++ b/CHANGELOG
@@ -1,7 +1,7 @@
 Development Version
 -------------------
 
-Nothing yet.
+* Fix CASE...END split-level tracking outside BEGIN blocks (pr839).
 
 
 Release 0.6.0 (Aug 13, 2026)
diff --git a/sqlparse/engine/statement_splitter.py b/sqlparse/engine/statement_splitter.py
index bc57d170..f691edcf 100644
--- a/sqlparse/engine/statement_splitter.py
+++ b/sqlparse/engine/statement_splitter.py
@@ -145,6 +145,10 @@ def _change_splitlevel(self, ttype, value):
             if res is not None:
                 return res
 
+        if unified == 'CASE':
+            self._block_stack.append('CASE')
+            return 1
+
         # Handle closing keywords
         return self._handle_closing_keyword(unified)
 
diff --git a/tests/test_split.py b/tests/test_split.py
index 92c3fefe..fec09e91 100644
--- a/tests/test_split.py
+++ b/tests/test_split.py
@@ -374,3 +374,44 @@ def test_split_standalone_for_update():
     assert stmts[1] == "SELECT 3;"
 
 
+def test_splitlevel_case_end():
+    # CASE in a plain SELECT did not increment the level, but its matching END
+    # decremented it unconditionally. This led to levels being wrong after the
+    # CASE WHEN ... END block.
+    s = sqlparse.engine.statement_splitter.StatementSplitter()
+    level = s.level
+
+    token_stream = [
+        (sqlparse.tokens.Keyword.DML, 'SELECT'),
+        (sqlparse.tokens.Keyword,     'CASE'),
+        (sqlparse.tokens.Keyword,     'WHEN'),
+        (sqlparse.tokens.Name,        'foo'),
+        (sqlparse.tokens.Keyword,     'THEN'),
+        (sqlparse.tokens.Number,      '1'),
+        (sqlparse.tokens.Keyword,     'END'),
+        (sqlparse.tokens.Keyword,     'FROM'),
+        (sqlparse.tokens.Name,        't'),
+    ]
+
+    for ttype, value in token_stream:
+        level += s._change_splitlevel(ttype, value)
+
+    assert level == 0
+
+    # This issue could lead to incorrectly treating a semicolon inside a text
+    # literal as a statement terminator and incorrectly splitting the query.
+    assert len(sqlparse.parse(
+        "SELECT CASE WHEN 1 THEN 2 END, test IN ('foo \\', 'foo;') FROM t"
+    )) == 1
+
+
+@pytest.mark.parametrize('expression', [
+    'CASE WHEN 1 THEN 2 END',
+    'CASE WHEN 1 THEN CASE WHEN 2 THEN 3 END ELSE 4 END',
+])
+def test_split_case_end_escaped_semicolon(expression):
+    query = "SELECT " + expression + ", test IN ('foo \\', 'foo;') FROM t"
+    statements = sqlparse.parse(query + '; SELECT 2;')
+    assert [str(statement) for statement in statements] == [
+        query + '; ', 'SELECT 2;',
+    ]