diff --git a/recommonmark/parser.py b/recommonmark/parser.py
index ec27fc0..9fb07f0 100644
--- a/recommonmark/parser.py
+++ b/recommonmark/parser.py
@@ -4,7 +4,6 @@ import itertools
 from docutils import parsers, nodes, utils
 
 from docutils.parsers.rst import roles, states
-from docutils.utils.code_analyzer import Lexer, LexerError
 
 from CommonMark import DocParser, HTMLRenderer
 from warnings import warn
@@ -123,32 +122,7 @@ class CommonMarkParser(parsers.Parser):
         self.current_node.append(verbatim_node)
 
     def code(self, language, text):
-        classes = ['code', 'highlight']
-
-        # embedded rst in codeblocks
-        if language == 'eval_rst':
-            rst_parser = parsers.rst.Parser()
-            rst_parser.parse(text, self.current_node.document)
-            return
-
-        if language:
-            classes.append(language)
-
-        try:
-            tokens = Lexer(utils.unescape(text, 1), language, True)
-        except LexerError as error:
-            raise error
-
-        node = nodes.literal_block(text, '', classes=classes)
-
-        # analyze content and add nodes for every token
-        for classes, value in tokens:
-            if classes:
-                node += nodes.inline(value, value, classes=classes)
-            else:
-                # insert as Text to decrease the verbosity of the output
-                node += nodes.Text(value, value)
-
+        node = nodes.literal_block(text, '', language=language)
         self.current_node.append(node)
 
     def paragraph(self, block):
diff --git a/recommonmark/transform.py b/recommonmark/transform.py
index e069242..b836cc0 100644
--- a/recommonmark/transform.py
+++ b/recommonmark/transform.py
@@ -3,6 +3,7 @@ import os
 import sys
 from states import DummyStateMachine
 from docutils import nodes, transforms
+from docutils.statemachine import StringList
 
 class AutoStructify(transforms.Transform):
     """Automatically try to transform blocks to sphinx directives.
@@ -162,27 +163,95 @@ class AutoStructify(transforms.Transform):
             node['refuri'] = uri
             return None
 
+    def auto_code_block(self, node):
+        """Try to automatically generate nodes for codeblock syntax.
+
+        Parameters
+        ----------
+        node : nodes.literal_block
+            Original codeblock node
+        Returns
+        -------
+        tocnode: docutils node
+            The converted toc tree node, None if conversion is not possible.
+        """
+
+        assert isinstance(node, nodes.literal_block)
+        if 'language' not in node:
+            return None
+        self.state_machine.reset(self.document,
+                                 node.parent,
+                                 self.current_level)
+        content = node.rawsource.split('\n')
+        language = node['language']
+        if language == 'math':
+            return self.state_machine.run_directive(
+            'math', content=content)
+        elif language == 'eval_rst':
+            # allow embed non section level rst
+            node = nodes.section()
+            self.state_machine.state.nested_parse(
+                StringList(content, source=node.rawsource),
+                0, node=node, match_titles=False)
+            return node.children[:]
+        # embedded rst in codeblocks
+        if language == 'eval_rst':
+            rst_parser = parsers.rst.Parser()
+            rst_parser.parse(text, self.current_node.document)
+            return
+            return None
+        else:
+            return self.state_machine.run_directive(
+                'code-block', arguments=[language],
+                content=content)
+
+    def find_replace(self, node):
+        """Try to find replace node for current node.
+
+        Parameters
+        ----------
+        node : docutil node
+            Node to find replacement for.
+
+        Returns
+        -------
+        nodes : node or list of node
+            The replacement nodes of current node.
+            Returns None if no replacement can be found.
+        """
+        newnode = None
+        if isinstance(node, nodes.Sequential):
+            newnode = self.auto_toc_tree(node)
+        elif isinstance(node, nodes.reference):
+            newnode = self.auto_doc_ref(node)
+        elif isinstance(node, nodes.literal_block):
+            newnode = self.auto_code_block(node)
+        return newnode
+
     def traverse(self, node):
         """Traverse the document tree rooted at node.
 
         node : docutil node
             current root node to traverse
         """
-        newnode = None
         old_level = self.current_level
         if isinstance(node, nodes.section):
             if 'level' in node:
                 self.current_level = node['level']
-        elif isinstance(node, nodes.Sequential):
-            newnode = self.auto_toc_tree(node)
-        elif isinstance(node, nodes.reference):
-            newnode = self.auto_doc_ref(node)
+        to_visit = []
+        to_replace = []
+        for c in node.children[:]:
+            newnode = self.find_replace(c)
+            if newnode is not None:
+                to_replace.append((c, newnode))
+            else:
+                to_visit.append(c)
 
-        if newnode:
-            node.replace_self(newnode)
-        else:
-            for c in node.children[:]:
-                self.traverse(c)
+        for oldnode, newnodes in to_replace:
+            node.replace(oldnode, newnodes)
+
+        for child in to_visit:
+            self.traverse(child)
         self.current_level = old_level
 
     def apply(self):