diff --git a/IPython/nbconvert/utils/lexers.py b/IPython/nbconvert/utils/lexers.py
index 99bd770..55aa070 100644
--- a/IPython/nbconvert/utils/lexers.py
+++ b/IPython/nbconvert/utils/lexers.py
@@ -53,6 +53,10 @@ ipython_tokens = [
                                        using(BashLexer), Text)),
   (r'(\%+)(\w+)\b', bygroups(Operator, Keyword)),
   (r'^(!)(.+)(\n)', bygroups(Operator, using(BashLexer), Text)),
+  (r'^(.+)(=)(\s*)(!)(.+)(\n)', bygroups(
+      # With the limited syntax allowed on the l.h.s. of a shell capture,
+      # we don't need to differentiate between Python 2 and 3.
+      using(Python3Lexer), Operator, Text, Operator, using(BashLexer), Text)),
 ]
 
 def build_ipy_lexer(python3):
diff --git a/IPython/nbconvert/utils/tests/test_lexers.py b/IPython/nbconvert/utils/tests/test_lexers.py
index d2371c1..b7a7ea4 100644
--- a/IPython/nbconvert/utils/tests/test_lexers.py
+++ b/IPython/nbconvert/utils/tests/test_lexers.py
@@ -33,3 +33,21 @@ class TestLexers(TestsBase):
             (Token.Text, '\n'),
         ]
         self.assertEqual(tokens, list(self.lexer.get_tokens(fragment)))
+
+        fragment_2 = 'x = ' + fragment
+        tokens_2 = [
+            (Token.Name, 'x'),
+            (Token.Text, ' '),
+            (Token.Operator, '='),
+            (Token.Text, ' '),
+        ] + tokens
+        self.assertEqual(tokens_2, list(self.lexer.get_tokens(fragment_2)))
+        fragment_2 = 'x, = ' + fragment
+        tokens_2 = [
+            (Token.Name, 'x'),
+            (Token.Punctuation, ','),
+            (Token.Text, ' '),
+            (Token.Operator, '='),
+            (Token.Text, ' '),
+        ] + tokens
+        self.assertEqual(tokens_2, list(self.lexer.get_tokens(fragment_2)))