diff --git a/docs/format.rst b/docs/format.rst
index 90129fb..984a7e9 100644
--- a/docs/format.rst
+++ b/docs/format.rst
@@ -25,5 +25,7 @@ The :mod:`ihm.format` Python module
 
 .. autoclass:: ChangeKeywordFilter
 
+.. autoclass:: ReplaceCategoryFilter
+
 .. autoexception:: CifParserError
    :members:
diff --git a/ihm/format.py b/ihm/format.py
index 172b361..fadacc2 100644
--- a/ihm/format.py
+++ b/ihm/format.py
@@ -14,6 +14,10 @@
 import textwrap
 import operator
 import ihm
+if sys.version_info[0] >= 3:
+    from io import StringIO
+else:
+    from io import BytesIO as StringIO
 # getargspec is deprecated in Python 3, but getfullargspec has a very
 # similar interface
 try:
@@ -666,6 +670,13 @@ def __init__(self, target):
             self.category = '_' + ts[0]
         self.keyword = ts[-1]
 
+    def _set_category_from_target(self, target):
+        if target.startswith('_'):
+            self.category = target
+        else:
+            self.category = '_' + target
+        self.keyword = None
+
     def match_token_category(self, tok):
         """Return true iff the given token matches the target's category"""
         return self.category is None or tok.category == self.category
@@ -684,6 +695,16 @@ def filter_category(self, tok):
         """
         raise NotImplementedError
 
+    def filter_loop_header(self, tok):
+        """Filter the given loop header token.
+
+           :return: the original token (which must not have been modified),
+                    a replacement token, or None if the token should be
+                    deleted. If the header token is replaced or deleted,
+                    all of the original loop rows will also be deleted.
+        """
+        return tok
+
     def get_loop_filter(self, tok):
         """Given a loop header token, potentially return a handler for each
            loop row token. This function is also permitted to alter the
@@ -791,6 +812,67 @@ def get_loop_filter(self, tok):
             tok.keywords[keyword_index].token.keyword = self.new
 
 
+class ReplaceCategoryFilter(Filter):
+    """Replace any token from the file that sets the given category.
+
+       This can also be used to completely remove a category if no
+       replacement is given.
+
+       :param str target: the mmCIF category name this filter should act on,
+              such as ``_entity``.
+       :param str raw_cif: if given, text in mmCIF format which should replace
+              the first instance of the category.
+       :param dumper: if given, a dumper object that should generate mmCIF
+              output to replace the first instance of the category.
+       :type dumper: :class:`ihm.dumper.Dumper`
+       :param system: the System that the given dumper will work on.
+       :type system: :class:`ihm.System`
+    """
+
+    class _RawCifToken(_Token):
+        __slots__ = ['txt']
+        category = keyword = None
+
+        def __init__(self, txt):
+            self.txt = txt
+
+        def as_mmcif(self):
+            return self.txt
+
+    def __init__(self, target, raw_cif=None, dumper=None, system=None):
+        self._set_category_from_target(target)
+        self.raw_cif = raw_cif
+        self.dumper = dumper
+        self.system = system
+        #: The number of times the category was found in the mmCIF file
+        self.num_matches = 0
+
+    def _get_replacement_token(self):
+        if self.num_matches > 1:
+            return None
+        if self.raw_cif:
+            return self._RawCifToken(self.raw_cif)
+        elif self.dumper and self.system:
+            fh = StringIO()
+            writer = CifWriter(fh)
+            self.dumper.finalize(self.system)
+            self.dumper.dump(self.system, writer)
+            return self._RawCifToken(fh.getvalue())
+
+    def filter_category(self, tok):
+        if self.match_token_category(tok):
+            self.num_matches += 1
+            return self._get_replacement_token()
+        else:
+            return tok
+
+    def filter_loop_header(self, tok):
+        return self.filter_category(tok)
+
+    def get_loop_filter(self, tok):
+        return None
+
+
 class CifTokenReader(_PreservingCifTokenizer):
     """Read an mmCIF file and break it into tokens.
 
@@ -836,13 +918,19 @@ def _read_file_with_filters(self, filters):
             if isinstance(tok, _CategoryTokenGroup):
                 tok = self._filter_category(tok, filters)
             elif isinstance(tok, ihm.format._LoopHeaderTokenGroup):
-                remove_all_loop_rows = False
-                loop_filters = [f.get_loop_filter(tok) for f in filters]
-                loop_filters = [f for f in loop_filters if f is not None]
-                # Did filters remove all keywords from the loop?
-                if all(isinstance(k.token, _NullToken) for k in tok.keywords):
-                    tok = None
+                new_tok = self._filter_loop_header(tok, filters)
+                if new_tok is not tok:
+                    tok = new_tok
                     remove_all_loop_rows = True
+                else:
+                    remove_all_loop_rows = False
+                    loop_filters = [f.get_loop_filter(tok) for f in filters]
+                    loop_filters = [f for f in loop_filters if f is not None]
+                    # Did filters remove all keywords from the loop?
+                    if all(isinstance(k.token, _NullToken)
+                           for k in tok.keywords):
+                        tok = None
+                        remove_all_loop_rows = True
             elif isinstance(tok, ihm.format._LoopRowTokenGroup):
                 if remove_all_loop_rows:
                     tok = None
@@ -858,6 +946,14 @@ def _filter_category(self, tok, filters):
                 return
         return tok
 
+    def _filter_loop_header(self, tok, filters):
+        orig_tok = tok
+        for f in filters:
+            tok = f.filter_loop_header(tok)
+            if tok is not orig_tok:
+                break
+        return tok
+
     def _filter_loop(self, tok, filters):
         for f in filters:
             tok = f(tok)
diff --git a/test/test_format.py b/test/test_format.py
index 4425da1..7faf961 100644
--- a/test/test_format.py
+++ b/test/test_format.py
@@ -15,6 +15,7 @@
 TOPDIR = os.path.abspath(os.path.join(os.path.dirname(__file__), '..'))
 utils.set_search_paths(TOPDIR)
 import ihm.format
+import ihm.dumper
 
 try:
     from ihm import _format
@@ -956,6 +957,46 @@ def test_cif_token_reader_filter(self):
 _foo.baz
 newa b c d
 x y
+""")
+
+    def test_cif_token_reader_replace_category_filter(self):
+        """Test CifTokenReader class with ReplaceCategoryFilter"""
+        cif = """
+data_foo_bar
+#
+_cat1.bar old
+#
+loop_
+_cat2.bar
+_cat2.baz
+a b c d
+x y
+#
+_cat3.x 1
+_cat3.y 2
+#
+_cat4.z 1
+"""
+        d = ihm.dumper._CommentDumper()
+        s = ihm.System()
+        s.comments.extend(['comment1', 'comment2'])
+        r = ihm.format.CifTokenReader(StringIO(cif))
+        filters = [ihm.format.ReplaceCategoryFilter("cat1"),
+                   ihm.format.ReplaceCategoryFilter("cat2", raw_cif='FOO'),
+                   ihm.format.ReplaceCategoryFilter("cat3", dumper=d,
+                                                    system=s)]
+        tokens = list(r.read_file(filters))
+        new_cif = "".join(x.as_mmcif() for x in tokens)
+        self.assertEqual(new_cif, """
+data_foo_bar
+#
+#
+FOO
+#
+# comment1
+# comment2
+#
+_cat4.z 1
 """)
 
     def test_category_token_group(self):