diff --git a/sphinx/application.py b/sphinx/application.py
index cee780c9c..c17cbb617 100644
--- a/sphinx/application.py
+++ b/sphinx/application.py
@@ -29,7 +29,7 @@ from docutils.parsers.rst import convert_directive_function, \
 import sphinx
 from sphinx import package_dir, locale
 from sphinx.roles import XRefRole
-from sphinx.config import Config
+from sphinx.config import Config, convert_numfig_format
 from sphinx.errors import SphinxError, SphinxWarning, ExtensionError, \
     VersionRequirementError, ConfigError
 from sphinx.domains import ObjType
@@ -236,6 +236,7 @@ class Sphinx(object):
         self._init_i18n()
         # check all configuration values for permissible types
         self.config.check_types(self.warn)
+        self._post_init_config()
         # set up source_parsers
         self._init_source_parsers()
         # set up the build environment
@@ -270,6 +271,9 @@ class Sphinx(object):
             else:
                 self.info('not available for built-in messages')
 
+    def _post_init_config(self):
+        convert_numfig_format(self.config)
+
     def _init_source_parsers(self):
         for suffix, parser in iteritems(self._additional_source_parsers):
             if suffix not in self.config.source_suffix:
diff --git a/sphinx/config.py b/sphinx/config.py
index 5741d66bf..1f1d6ec30 100644
--- a/sphinx/config.py
+++ b/sphinx/config.py
@@ -286,3 +286,12 @@ class Config(object):
 
     def __contains__(self, name):
         return name in self.values
+
+
+def convert_numfig_format(config):
+    """ Convert numfi_format to new one. """
+
+    numfig_format = {}
+    for figtype, fmt in iteritems(config.numfig_format):
+        numfig_format[figtype] = fmt.replace('%s', '{number}')
+    config.numfig_format = numfig_format
diff --git a/sphinx/domains/std.py b/sphinx/domains/std.py
index b7f2597d4..f83afdab1 100644
--- a/sphinx/domains/std.py
+++ b/sphinx/domains/std.py
@@ -673,20 +673,18 @@ class StandardDomain(Domain):
             else:
                 title = env.config.numfig_format.get(figtype, '')
 
-            if figname is None and '%{name}' in title:
+            # convert old styled numfig_format to new style
+            title = title.replace('%s', '{number}')
+
+            if figname is None and '{name}' in title:
                 env.warn_node('the link has no caption: %s' % title, node)
                 return contnode
             else:
                 fignum = '.'.join(map(str, fignumber))
-                if '{name}' in title or 'number' in title:
-                    # new style format (cf. "Fig.%{number}")
-                    if figname:
-                        newtitle = title.format(name=figname, number=fignum)
-                    else:
-                        newtitle = title.format(number=fignum)
+                if figname:
+                    newtitle = title.format(name=figname, number=fignum)
                 else:
-                    # old style format (cf. "Fig.%s")
-                    newtitle = title % fignum
+                    newtitle = title.format(number=fignum)
         except KeyError as exc:
             env.warn_node('invalid numfig_format: %s (%r)' % (title, exc), node)
             return contnode
diff --git a/sphinx/util/texescape.py b/sphinx/util/texescape.py
index 0a3192f6a..861026b55 100644
--- a/sphinx/util/texescape.py
+++ b/sphinx/util/texescape.py
@@ -118,10 +118,17 @@ tex_replacements = [
     ('Ω', r'\(\Omega\)'),
     ('Ω', r'\(\Omega\)'),
 ]
+tex_pyformats = [
+    # map special chars of str.format() to escaped one
+    ('{', '{{'),
+    ('}', '}}'),
+]
+
 
 tex_escape_map = {}
 tex_replace_map = {}
 tex_hl_escape_map_new = {}
+tex_pyformat_map = {}
 
 
 def init():
@@ -133,3 +140,6 @@ def init():
         if a in '[]{}\\':
             continue
         tex_hl_escape_map_new[ord(a)] = b
+
+    for a, b in tex_pyformats:
+        tex_pyformat_map[ord(a)] = b
diff --git a/sphinx/writers/html.py b/sphinx/writers/html.py
index ba2b758d8..7dd286cf7 100644
--- a/sphinx/writers/html.py
+++ b/sphinx/writers/html.py
@@ -291,7 +291,7 @@ class HTMLTranslator(BaseTranslator):
                     self.builder.warn(msg)
                 else:
                     numbers = self.builder.fignumbers[key][figure_id]
-                    self.body.append(prefix % '.'.join(map(str, numbers)) + ' ')
+                    self.body.append(prefix.format(number='.'.join(map(str, numbers))) + ' ')
                     self.body.append('</span>')
 
         figtype = self.builder.env.domains['std'].get_figtype(node)
diff --git a/sphinx/writers/latex.py b/sphinx/writers/latex.py
index ca0f1a7dd..0a158297b 100644
--- a/sphinx/writers/latex.py
+++ b/sphinx/writers/latex.py
@@ -29,7 +29,7 @@ from sphinx.util import split_into
 from sphinx.util.i18n import format_date
 from sphinx.util.nodes import clean_astext, traverse_parent
 from sphinx.util.template import LaTeXRenderer
-from sphinx.util.texescape import tex_escape_map, tex_replace_map
+from sphinx.util.texescape import tex_escape_map, tex_replace_map, tex_pyformat_map
 from sphinx.util.smartypants import educate_quotes_latex
 
 
@@ -610,7 +610,7 @@ class LaTeXTranslator(nodes.NodeVisitor):
 
     def generate_numfig_format(self, builder):
         ret = []
-        figure = self.builder.config.numfig_format['figure'].split('%s', 1)
+        figure = self.builder.config.numfig_format['figure'].split('{number}', 1)
         if len(figure) == 1:
             ret.append('\\def\\fnum@figure{%s}\n' %
                        escape_abbr(text_type(figure[0]).translate(tex_escape_map)))
@@ -623,7 +623,7 @@ class LaTeXTranslator(nodes.NodeVisitor):
                            escape_abbr(text_type(figure[1]).translate(tex_escape_map)))
                 ret.append('\\makeatother\n')
 
-        table = self.builder.config.numfig_format['table'].split('%s', 1)
+        table = self.builder.config.numfig_format['table'].split('{number}', 1)
         if len(table) == 1:
             ret.append('\\def\\fnum@table{%s}\n' %
                        escape_abbr(text_type(table[0]).translate(tex_escape_map)))
@@ -636,7 +636,7 @@ class LaTeXTranslator(nodes.NodeVisitor):
                            escape_abbr(text_type(table[1]).translate(tex_escape_map)))
                 ret.append('\\makeatother\n')
 
-        codeblock = self.builder.config.numfig_format['code-block'].split('%s', 1)
+        codeblock = self.builder.config.numfig_format['code-block'].split('{number}', 1)
         if len(codeblock) == 1:
             pass  # FIXME
         else:
@@ -1789,16 +1789,11 @@ class LaTeXTranslator(nodes.NodeVisitor):
         else:
             id = node.get('refuri', '')[1:].replace('#', ':')
 
-        title = node.get('title', '%s')
-        title = text_type(title).translate(tex_escape_map).replace('\\%s', '%s')
-        if '\\{name\\}' in title or '\\{number\\}' in title:
-            # new style format (cf. "Fig.%{number}")
-            title = title.replace('\\{name\\}', '{name}').replace('\\{number\\}', '{number}')
-            text = escape_abbr(title).format(name='\\nameref{%s}' % self.idescape(id),
-                                             number='\\ref{%s}' % self.idescape(id))
-        else:
-            # old style format (cf. "Fig.%{number}")
-            text = escape_abbr(title) % ('\\ref{%s}' % self.idescape(id))
+        title = node.get('title', '{number}').replace('%s', '{number}')
+        title = text_type(title).translate(tex_escape_map).translate(tex_pyformat_map)
+        title = title.replace('\\{{name\\}}', '{name}').replace('\\{{number\\}}', '{number}')
+        text = escape_abbr(title).format(name='\\nameref{%s}' % self.idescape(id),
+                                         number='\\ref{%s}' % self.idescape(id))
         hyperref = '\\hyperref[%s]{%s}' % (self.idescape(id), text)
         self.body.append(hyperref)
 
diff --git a/tests/roots/test-numfig/index.rst b/tests/roots/test-numfig/index.rst
index 939903839..499a1b7dc 100644
--- a/tests/roots/test-numfig/index.rst
+++ b/tests/roots/test-numfig/index.rst
@@ -53,7 +53,5 @@ test-tocdepth
 * Section.1 is :numref:`foo`
 * Section.2.1 is :numref:`bar_a`
 * Unnumbered section is :numref:`index`
-* Invalid numfig_format 01: :numref:`invalid <fig1>`
-* Invalid numfig_format 02: :numref:`Fig %s %s <fig1>`
 * Fig.1 is :numref:`Fig.{number} {name} <fig1>`
 * Section.1 is :numref:`Sect.{number} {name} <foo>`
diff --git a/tests/test_build_html.py b/tests/test_build_html.py
index d8aff88ab..176590f75 100644
--- a/tests/test_build_html.py
+++ b/tests/test_build_html.py
@@ -506,8 +506,6 @@ def test_numfig_disabled(app, status, warning):
     warnings = warning.getvalue()
     assert 'index.rst:47: WARNING: numfig is disabled. :numref: is ignored.' in warnings
     assert 'index.rst:55: WARNING: no number is assigned for section: index' not in warnings
-    assert 'index.rst:56: WARNING: invalid numfig_format: invalid' not in warnings
-    assert 'index.rst:57: WARNING: invalid numfig_format: Fig %s %s' not in warnings
 
     expects = {
         'index.html': [
@@ -570,8 +568,6 @@ def test_numfig_without_numbered_toctree(app, status, warning):
     warnings = warning.getvalue()
     assert 'index.rst:47: WARNING: numfig is disabled. :numref: is ignored.' not in warnings
     assert 'index.rst:55: WARNING: no number is assigned for section: index' in warnings
-    assert 'index.rst:56: WARNING: invalid numfig_format: invalid' in warnings
-    assert 'index.rst:57: WARNING: invalid numfig_format: Fig %s %s' in warnings
 
     expects = {
         'index.html': [
@@ -670,8 +666,6 @@ def test_numfig_with_numbered_toctree(app, status, warning):
     warnings = warning.getvalue()
     assert 'index.rst:47: WARNING: numfig is disabled. :numref: is ignored.' not in warnings
     assert 'index.rst:55: WARNING: no number is assigned for section: index' in warnings
-    assert 'index.rst:56: WARNING: invalid numfig_format: invalid' in warnings
-    assert 'index.rst:57: WARNING: invalid numfig_format: Fig %s %s' in warnings
 
     expects = {
         'index.html': [
@@ -774,8 +768,6 @@ def test_numfig_with_prefix(app, status, warning):
     warnings = warning.getvalue()
     assert 'index.rst:47: WARNING: numfig is disabled. :numref: is ignored.' not in warnings
     assert 'index.rst:55: WARNING: no number is assigned for section: index' in warnings
-    assert 'index.rst:56: WARNING: invalid numfig_format: invalid' in warnings
-    assert 'index.rst:57: WARNING: invalid numfig_format: Fig %s %s' in warnings
 
     expects = {
         'index.html': [
@@ -874,8 +866,6 @@ def test_numfig_with_secnum_depth(app, status, warning):
     warnings = warning.getvalue()
     assert 'index.rst:47: WARNING: numfig is disabled. :numref: is ignored.' not in warnings
     assert 'index.rst:55: WARNING: no number is assigned for section: index' in warnings
-    assert 'index.rst:56: WARNING: invalid numfig_format: invalid' in warnings
-    assert 'index.rst:57: WARNING: invalid numfig_format: Fig %s %s' in warnings
 
     expects = {
         'index.html': [