diff --git a/libmproxy/console.py b/libmproxy/console.py
index d99dd8ac5..ffe37fc31 100644
--- a/libmproxy/console.py
+++ b/libmproxy/console.py
@@ -93,7 +93,7 @@ def format_flow(f, focus, extended=False, padding=2):
             txt.append(("goodcode", str(f.response.code)))
         else:
             txt.append(("error", str(f.response.code)))
-        t = f.response.headers.get("content-type")
+        t = f.response.headers["content-type"]
         if t:
             t = t[0].split(";")[0]
             txt.append(("text", " %s"%t))
@@ -295,7 +295,11 @@ class ConnectionView(WWrap):
 
     def _conn_text(self, conn, viewmode):
         if conn:
-            return self.master._cached_conn_text(conn.content, tuple(conn.headers.itemPairs()), viewmode)
+            return self.master._cached_conn_text(
+                        conn.content,
+                        tuple([tuple(i) for i in conn.headers.lst]),
+                        viewmode
+                    )
         else:
             return urwid.ListBox([])
 
@@ -485,7 +489,7 @@ class ConnectionView(WWrap):
             else:
                 conn = self.flow.response
             if conn.content:
-                t = conn.headers.get("content-type", [None])
+                t = conn.headers["content-type"] or [None]
                 t = t[0]
                 self.master.spawn_external_viewer(conn.content, t)
         elif key == "b":
diff --git a/libmproxy/filt.py b/libmproxy/filt.py
index 31c43581b..40cf7358b 100644
--- a/libmproxy/filt.py
+++ b/libmproxy/filt.py
@@ -79,7 +79,7 @@ class _Rex(_Action):
             raise ValueError, "Cannot compile expression."
 
 def _check_content_type(expr, o):
-    val = o.headers.get("content-type")
+    val = o.headers["content-type"]
     if val and re.search(expr, val[0]):
         return True
     return False
diff --git a/libmproxy/flow.py b/libmproxy/flow.py
index be77af337..d29b8e2d8 100644
--- a/libmproxy/flow.py
+++ b/libmproxy/flow.py
@@ -101,7 +101,7 @@ class ServerPlaybackState:
         if self.headers:
             hdrs = []
             for i in self.headers:
-                v = r.headers.get(i, [])
+                v = r.headers[i]
                 # Slightly subtle: we need to convert everything to strings
                 # to prevent a mismatch between unicode/non-unicode.
                 v = [str(x) for x in v]
@@ -139,7 +139,7 @@ class StickyCookieState:
         )
 
     def handle_response(self, f):
-        for i in f.response.headers.get("set-cookie", []):
+        for i in f.response.headers["set-cookie"]:
             # FIXME: We now know that Cookie.py screws up some cookies with
             # valid RFC 822/1123 datetime specifications for expiry. Sigh.
             c = Cookie.SimpleCookie(i)
@@ -158,9 +158,10 @@ class StickyCookieState:
                     f.request.path.startswith(i[2])
                 ]
                 if all(match):
-                    l = f.request.headers.setdefault("cookie", [])
+                    l = f.request.headers["cookie"]
                     f.request.stickycookie = True
                     l.append(self.jar[i].output(header="").strip())
+                    f.request.headers["cookie"] = l
 
 
 class StickyAuthState:
diff --git a/libmproxy/proxy.py b/libmproxy/proxy.py
index 362d622dd..690df9f44 100644
--- a/libmproxy/proxy.py
+++ b/libmproxy/proxy.py
@@ -56,11 +56,11 @@ def read_chunked(fp):
     
 
 def read_http_body(rfile, connection, headers, all):
-    if headers.has_key('transfer-encoding'):
+    if 'transfer-encoding' in headers:
         if not ",".join(headers["transfer-encoding"]) == "chunked":
             raise IOError('Invalid transfer-encoding')
         content = read_chunked(rfile)
-    elif headers.has_key("content-length"):
+    elif "content-length" in headers:
         content = rfile.read(int(headers["content-length"][0]))
     elif all:
         content = rfile.read()
@@ -152,8 +152,7 @@ class Request(controller.Msg):
             "if-none-match",
         ]
         for i in delheaders:
-            if i in self.headers:
-                del self.headers[i]
+            del self.headers[i]
 
     def set_replay(self):
         self.client_conn = None
@@ -251,7 +250,7 @@ class Request(controller.Msg):
         utils.try_del(headers, 'connection')
         utils.try_del(headers, 'content-length')
         utils.try_del(headers, 'transfer-encoding')
-        if not headers.has_key('host'):
+        if not 'host' in headers:
             headers["host"] = [self.hostport()]
         content = self.content
         if content is not None:
@@ -321,7 +320,7 @@ class Response(controller.Msg):
                     new = mktime_tz(d) + delta
                     self.headers[i] = [formatdate(new)]
         c = []
-        for i in self.headers.get("set-cookie", []):
+        for i in self.headers["set-cookie"]:
             c.append(self._refresh_cookie(i, delta))
         if c:
             self.headers["set-cookie"] = c
@@ -656,7 +655,7 @@ class ProxyHandler(SocketServer.StreamRequestHandler):
             scheme = "https"
         headers = utils.Headers()
         headers.read(self.rfile)
-        if host is None and headers.has_key("host"):
+        if host is None and "host" in headers:
             netloc = headers["host"][0]
             if ':' in netloc:
                 host, port = string.split(netloc, ':')
@@ -670,7 +669,7 @@ class ProxyHandler(SocketServer.StreamRequestHandler):
             port = int(port)
         if host is None:
             raise ProxyError(400, 'Invalid request: %s'%request)
-        if headers.has_key('expect'):
+        if "expect" in headers:
             expect = ",".join(headers['expect'])
             if expect == "100-continue" and httpminor >= 1:
                 self.wfile.write('HTTP/1.1 100 Continue\r\n')
@@ -681,7 +680,7 @@ class ProxyHandler(SocketServer.StreamRequestHandler):
                 raise ProxyError(417, 'Unmet expect: %s'%expect)
         if httpminor == 0:
             client_conn.close = True
-        if headers.has_key('connection'):
+        if "connection" in headers:
             for value in ",".join(headers['connection']).split(","):
                 value = value.strip()
                 if value == "close":
diff --git a/libmproxy/utils.py b/libmproxy/utils.py
index 8ac1f5470..38fc61078 100644
--- a/libmproxy/utils.py
+++ b/libmproxy/utils.py
@@ -12,7 +12,7 @@
 # 
 # You should have received a copy of the GNU General Public License
 # along with this program.  If not, see <http://www.gnu.org/licenses/>.
-import re, os, subprocess, datetime, textwrap, errno, sys, time, functools
+import re, os, subprocess, datetime, textwrap, errno, sys, time, functools, copy
 import json
 
 CERT_SLEEP_TIME = 1
@@ -164,10 +164,6 @@ def isSequenceLike(anobj):
     return 1
 
 
-def _caseless(s):
-    return s.lower()
-
-
 def try_del(dict, key):
     try:
         del dict[key]
@@ -175,108 +171,72 @@ def try_del(dict, key):
         pass
 
 
-class MultiDict:
-    """
-        Simple wrapper around a dictionary to make holding multiple objects per
-        key easier.
+class Headers:
+    def __init__(self, lst=None):
+        if lst:
+            self.lst = lst
+        else:
+            self.lst = []
 
-        Note that this class assumes that keys are strings.
-
-        Keys have no order, but the order in which values are added to a key is
-        preserved.
-    """
-    # This ridiculous bit of subterfuge is needed to prevent the class from
-    # treating this as a bound method.
-    _helper = (str,)
-    def __init__(self):
-        self._d = dict()
-
-    def copy(self):
-        m = self.__class__()
-        m._d = self._d.copy()
-        return m
-
-    def clear(self):
-        return self._d.clear()
-
-    def get(self, key, d=None):
-        key = self._helper[0](key)
-        return self._d.get(key, d)
-
-    def __contains__(self, key):
-        key = self._helper[0](key)
-        return self._d.__contains__(key)
+    def _kconv(self, s):
+        return s.lower()
 
     def __eq__(self, other):
-        return dict(self) == dict(other)
+        return self.lst == other.lst
 
-    def __delitem__(self, key):
-        self._d.__delitem__(key)
+    def __getitem__(self, k):
+        ret = []
+        k = self._kconv(k)
+        for i in self.lst:
+            if self._kconv(i[0]) == k:
+                ret.append(i[1])
+        return ret
 
-    def __getitem__(self, key):
-        key = self._helper[0](key)
-        return self._d.__getitem__(key)
-    
-    def __setitem__(self, key, value):
-        if not isSequenceLike(value):
-            raise ValueError, "Cannot insert non-sequence."
-        key = self._helper[0](key)
-        return self._d.__setitem__(key, value)
+    def _filter_lst(self, k, lst):
+        new = []
+        for i in lst:
+            if self._kconv(i[0]) != k:
+                new.append(i)
+        return new
 
-    def has_key(self, key):
-        key = self._helper[0](key)
-        return self._d.has_key(key)
+    def __setitem__(self, k, hdrs):
+        k = self._kconv(k)
+        first = None
+        new = self._filter_lst(k, self.lst)
+        for i in hdrs:
+            new.append((k, i))
+        self.lst = new
 
-    def setdefault(self, key, default=None):
-        key = self._helper[0](key)
-        return self._d.setdefault(key, default)
+    def __delitem__(self, k):
+        self.lst = self._filter_lst(k, self.lst)
 
-    def keys(self):
-        return self._d.keys()
+    def __contains__(self, k):
+        for i in self.lst:
+            if self._kconv(i[0]) == k:
+                return True
+        return False
 
-    def extend(self, key, value):
-        if not self.has_key(key):
-            self[key] = []
-        self[key].extend(value)
-
-    def append(self, key, value):
-        self.extend(key, [value])
-
-    def itemPairs(self):
-        """
-            Yield all possible pairs of items.
-        """
-        for i in self.keys():
-            for j in self[i]:
-                yield (i, j)
+    def add(self, key, value):
+        self.lst.append([key, str(value)])
 
     def get_state(self):
-        return list(self.itemPairs())
+        return [tuple(i) for i in self.lst]
 
     @classmethod
     def from_state(klass, state):
-        md = klass()
-        for i in state:
-            md.append(*i)
-        return md
+        return klass([list(i) for i in state])
 
+    def copy(self):
+        lst = copy.deepcopy(self.lst)
+        return Headers(lst)
 
-class Headers(MultiDict):
-    """
-        A dictionary-like class for keeping track of HTTP headers.
-
-        It is case insensitive, and __repr__ formats the headers correcty for
-        output to the server.
-    """
-    _helper = (_caseless,)
     def __repr__(self):
         """
             Returns a string containing a formatted header string.
         """
         headerElements = []
-        for key in sorted(self.keys()):
-            for val in self[key]:
-                headerElements.append(key + ": " + val)
+        for itm in self.lst:
+            headerElements.append(itm[0] + ": " + itm[1])
         headerElements.append("")
         return "\r\n".join(headerElements)
 
@@ -284,7 +244,7 @@ class Headers(MultiDict):
         """
             Match the regular expression against each header (key, value) pair.
         """
-        for k, v in self.itemPairs():
+        for k, v in self.lst:
             s = "%s: %s"%(k, v)
             if re.search(expr, s):
                 return True
@@ -295,6 +255,7 @@ class Headers(MultiDict):
             Read a set of headers from a file pointer. Stop once a blank line
             is reached.
         """
+        ret = []
         name = ''
         while 1:
             line = fp.readline()
@@ -302,18 +263,15 @@ class Headers(MultiDict):
                 break
             if line[0] in ' \t':
                 # continued header
-                self[name][-1] = self[name][-1] + '\r\n ' + line.strip()
+                ret[-1][1] = ret[-1][1] + '\r\n ' + line.strip()
             else:
                 i = line.find(':')
                 # We're being liberal in what we accept, here.
                 if i > 0:
                     name = line[:i]
                     value = line[i+1:].strip()
-                    if self.has_key(name):
-                        # merge value
-                        self.append(name, value)
-                    else:
-                        self[name] = [value]
+                    ret.append([name, value])
+        self.lst = ret
 
 
 def pretty_size(size):
diff --git a/test/test_server.py b/test/test_server.py
index e9b611657..1e3c1df49 100644
--- a/test/test_server.py
+++ b/test/test_server.py
@@ -44,7 +44,7 @@ class uProxy(tutils.ProxTest):
             
         l = self.log()
         assert l[0].address
-        assert l[1].headers.has_key("host")
+        assert "host" in l[1].headers
         assert l[2].code == 200
 
     def test_https(self):
@@ -55,7 +55,7 @@ class uProxy(tutils.ProxTest):
 
         l = self.log()
         assert l[0].address
-        assert l[1].headers.has_key("host")
+        assert "host" in l[1].headers
         assert l[2].code == 200
 
     # Disable these two for now: they take a long time.
diff --git a/test/test_utils.py b/test/test_utils.py
index b64db918a..d5957872d 100644
--- a/test/test_utils.py
+++ b/test/test_utils.py
@@ -43,88 +43,6 @@ class uData(libpry.AutoTree):
         libpry.raises("does not exist", utils.data.path, "nonexistent")
 
 
-class uMultiDict(libpry.AutoTree):
-    def setUp(self):
-        self.md = utils.MultiDict()
-
-    def test_setget(self):
-        assert not self.md.has_key("foo")
-        self.md.append("foo", 1)
-        assert self.md["foo"] == [1]
-        assert self.md.has_key("foo")
-
-    def test_del(self):
-        self.md.append("foo", 1)
-        del self.md["foo"]
-        assert not self.md.has_key("foo")
-
-    def test_extend(self):
-        self.md.append("foo", 1)
-        self.md.extend("foo", [2, 3])
-        assert self.md["foo"] == [1, 2, 3]
-
-    def test_extend_err(self):
-        self.md.append("foo", 1)
-        libpry.raises("not iterable", self.md.extend, "foo", 2)
-
-    def test_get(self):
-        self.md.append("foo", 1)
-        self.md.append("foo", 2)
-        assert self.md.get("foo") == [1, 2]
-        assert self.md.get("bar") == None
-
-    def test_caseSensitivity(self):
-        self.md._helper = (utils._caseless,)
-        self.md["foo"] = [1]
-        self.md.append("FOO", 2)
-        assert self.md["foo"] == [1, 2]
-        assert self.md["FOO"] == [1, 2]
-        assert self.md.has_key("FoO")
-
-    def test_dict(self):
-        self.md.append("foo", 1)
-        self.md.append("foo", 2)
-        self.md["bar"] = [3]
-        assert self.md == self.md
-        assert dict(self.md) == self.md
-
-    def test_copy(self):
-        self.md["foo"] = [1, 2]
-        self.md["bar"] = [3, 4]
-        md2 = self.md.copy()
-        assert md2 == self.md
-        assert id(md2) != id(self.md)
-
-    def test_clear(self):
-        self.md["foo"] = [1, 2]
-        self.md["bar"] = [3, 4]
-        self.md.clear()
-        assert not self.md.keys()
-
-    def test_setitem(self):
-        libpry.raises(ValueError, self.md.__setitem__, "foo", "bar")
-        self.md["foo"] = ["bar"]
-        assert self.md["foo"] == ["bar"]
-
-    def test_itemPairs(self):
-        self.md.append("foo", 1)
-        self.md.append("foo", 2)
-        self.md.append("bar", 3)
-        l = list(self.md.itemPairs())
-        assert len(l) == 3
-        assert ("foo", 1) in l
-        assert ("foo", 2) in l
-        assert ("bar", 3) in l
-
-    def test_getset_state(self):
-        self.md.append("foo", 1)
-        self.md.append("foo", 2)
-        self.md.append("bar", 3)
-        state = self.md.get_state()
-        nd = utils.MultiDict.from_state(state)
-        assert nd == self.md
-
-
 class uHeaders(libpry.AutoTree):
     def setUp(self):
         self.hd = utils.Headers()
@@ -168,9 +86,9 @@ class uHeaders(libpry.AutoTree):
         assert self.hd["header"] == ['one\r\n two']
 
     def test_dictToHeader1(self):
-        self.hd.append("one", "uno")
-        self.hd.append("two", "due")
-        self.hd.append("two", "tre")
+        self.hd.add("one", "uno")
+        self.hd.add("two", "due")
+        self.hd.add("two", "tre")
         expected = [
             "one: uno\r\n",
             "two: due\r\n",
@@ -191,21 +109,34 @@ class uHeaders(libpry.AutoTree):
 
     def test_match_re(self):
         h = utils.Headers()
-        h.append("one", "uno")
-        h.append("two", "due")
-        h.append("two", "tre")
+        h.add("one", "uno")
+        h.add("two", "due")
+        h.add("two", "tre")
         assert h.match_re("uno")
         assert h.match_re("two: due")
         assert not h.match_re("nonono")
 
     def test_getset_state(self):
-        self.hd.append("foo", 1)
-        self.hd.append("foo", 2)
-        self.hd.append("bar", 3)
+        self.hd.add("foo", 1)
+        self.hd.add("foo", 2)
+        self.hd.add("bar", 3)
         state = self.hd.get_state()
         nd = utils.Headers.from_state(state)
         assert nd == self.hd
 
+    def test_copy(self):
+        self.hd.add("foo", 1)
+        self.hd.add("foo", 2)
+        self.hd.add("bar", 3)
+        assert self.hd == self.hd.copy()
+
+    def test_del(self):
+        self.hd.add("foo", 1)
+        self.hd.add("Foo", 2)
+        self.hd.add("bar", 3)
+        del self.hd["foo"]
+        assert len(self.hd.lst) == 1
+
 
 class uisStringLike(libpry.AutoTree):
     def test_all(self):
@@ -371,7 +302,6 @@ tests = [
     upretty_size(),
     uisStringLike(),
     uisSequenceLike(),
-    uMultiDict(),
     uHeaders(),
     uData(),
     upretty_xmlish(),