diff --git a/include/boost/thread.hpp b/include/boost/thread.hpp
index b9a9ed68..625e69ec 100644
--- a/include/boost/thread.hpp
+++ b/include/boost/thread.hpp
@@ -4,6 +4,8 @@
 // Distributed under the Boost Software License, Version 1.0. (See accompanying 
 // file LICENSE_1_0.txt or copy at http://www.boost.org/LICENSE_1_0.txt)
 
+//  See www.boost.org/libs/thread for documentation.
+
 #if !defined(BOOST_THREAD_WEK01082003_HPP)
 #define BOOST_THREAD_WEK01082003_HPP
 
diff --git a/include/boost/thread/win32/thread_heap_alloc.hpp b/include/boost/thread/win32/thread_heap_alloc.hpp
index 1835e35e..ee47e0e6 100644
--- a/include/boost/thread/win32/thread_heap_alloc.hpp
+++ b/include/boost/thread/win32/thread_heap_alloc.hpp
@@ -156,25 +156,13 @@ namespace boost
         }
 
         template<typename T>
-        struct do_delete
+        struct do_heap_delete
         {
-            T* data;
-            
-            do_delete(T* data_):
-                data(data_)
-            {}
-            
-            void operator()() const
+            void operator()(T* data) const
             {
                 detail::heap_delete(data);
             }
         };
-
-        template<typename T>
-        do_delete<T> make_heap_deleter(T* data)
-        {
-            return do_delete<T>(data);
-        }
     }
 }
 
diff --git a/include/boost/thread/win32/tss.hpp b/include/boost/thread/win32/tss.hpp
index a99dfac4..fcf792bd 100644
--- a/include/boost/thread/win32/tss.hpp
+++ b/include/boost/thread/win32/tss.hpp
@@ -1,13 +1,26 @@
 #ifndef BOOST_THREAD_WIN32_TSS_HPP
 #define BOOST_THREAD_WIN32_TSS_HPP
+// Distributed under the Boost Software License, Version 1.0. (See
+// accompanying file LICENSE_1_0.txt or copy at
+// http://www.boost.org/LICENSE_1_0.txt)
+// (C) Copyright 2007 Anthony Williams
+
+#include <boost/shared_ptr.hpp>
+#include "thread_heap_alloc.hpp"
 
 namespace boost
 {
     namespace detail
     {
-        typedef void(*tss_cleanup_function)(void const* key,void* value);
+        struct tss_cleanup_function
+        {
+            virtual ~tss_cleanup_function()
+            {}
+            
+            virtual void operator()(void* data)=0;
+        };
         
-        BOOST_THREAD_DECL void set_tss_data(void const* key,tss_cleanup_function func,void* tss_data,bool cleanup_existing);
+        BOOST_THREAD_DECL void set_tss_data(void const* key,boost::shared_ptr<tss_cleanup_function> func,void* tss_data,bool cleanup_existing);
         BOOST_THREAD_DECL void* get_tss_data(void const* key);
     }
 
@@ -18,31 +31,39 @@ namespace boost
         thread_specific_ptr(thread_specific_ptr&);
         thread_specific_ptr& operator=(thread_specific_ptr&);
 
-        static void delete_data(void const* self,void* value)
+        struct delete_data:
+            detail::tss_cleanup_function
         {
-            static_cast<thread_specific_ptr const*>(self)->cleanup((T*)value);
-        }
+            void operator()(void* data)
+            {
+                delete static_cast<T*>(data);
+            }
+        };
         
-        void cleanup(T* data) const
+        struct run_custom_cleanup_function:
+            detail::tss_cleanup_function
         {
-            if(func)
+            void (*cleanup_function)(T*);
+            
+            explicit run_custom_cleanup_function(void (*cleanup_function_)(T*)):
+                cleanup_function(cleanup_function_)
+            {}
+            
+            void operator()(void* data)
             {
-                func(data);
+                cleanup_function(data);
             }
-            else
-            {
-                delete data;
-            }
-        }
+        };
 
-        void (*func)(T*);
+
+        boost::shared_ptr<detail::tss_cleanup_function> cleanup;
         
     public:
         thread_specific_ptr():
-            func(0)
+            cleanup(detail::heap_new<delete_data>(),detail::do_heap_delete<delete_data>())
         {}
         explicit thread_specific_ptr(void (*func_)(T*)):
-            func(func_)
+            cleanup(detail::heap_new<run_custom_cleanup_function>(func_),detail::do_heap_delete<run_custom_cleanup_function>())
         {}
         ~thread_specific_ptr()
         {
@@ -72,7 +93,7 @@ namespace boost
             T* const current_value=get();
             if(current_value!=new_value)
             {
-                detail::set_tss_data(this,delete_data,new_value,true);
+                detail::set_tss_data(this,cleanup,new_value,true);
             }
         }
     };
diff --git a/src/win32/thread.cpp b/src/win32/thread.cpp
index fee2b4fd..9df6a05a 100644
--- a/src/win32/thread.cpp
+++ b/src/win32/thread.cpp
@@ -20,28 +20,6 @@ namespace boost
 {
     namespace
     {
-#if defined(_MSC_VER) && !defined(UNDER_CE)
-        __declspec(thread) detail::thread_data_base* current_thread_data=0;
-        detail::thread_data_base* get_current_thread_data()
-        {
-            return current_thread_data;
-        }
-        void set_current_thread_data(detail::thread_data_base* new_data)
-        {
-            current_thread_data=new_data;
-        }
-#elif defined(__BORLANDC__)
-        detail::thread_data_base* __thread current_thread_data=0;
-        detail::thread_data_base* get_current_thread_data()
-        {
-            return current_thread_data;
-        }
-        void set_current_thread_data(detail::thread_data_base* new_data)
-        {
-            current_thread_data=new_data;
-        }
-#else
-
         boost::once_flag current_thread_tls_init_flag=BOOST_ONCE_INIT;
         DWORD current_thread_tls_key=0;
 
@@ -62,7 +40,6 @@ namespace boost
             boost::call_once(current_thread_tls_init_flag,create_current_thread_tls_key);
             BOOST_VERIFY(TlsSetValue(current_thread_tls_key,new_data));
         }
-#endif
 
 #ifdef BOOST_NO_THREADEX
 // Windows CE doesn't define _beginthreadex
@@ -135,11 +112,11 @@ namespace boost
         struct tss_data_node
         {
             void const* key;
-            boost::detail::tss_cleanup_function func;
+            boost::shared_ptr<boost::detail::tss_cleanup_function> func;
             void* value;
             tss_data_node* next;
 
-            tss_data_node(void const* key_,boost::detail::tss_cleanup_function func_,void* value_,
+            tss_data_node(void const* key_,boost::shared_ptr<boost::detail::tss_cleanup_function> func_,void* value_,
                           tss_data_node* next_):
                 key(key_),func(func_),value(value_),next(next_)
             {}
@@ -173,7 +150,7 @@ namespace boost
                         current_thread_data->tss_data=current_node->next;
                         if(current_node->func)
                         {
-                            (*current_node->func)(current_node->key,current_node->value);
+                            (*current_node->func)(current_node->value);
                         }
                         boost::detail::heap_delete(current_node);
                     }
@@ -489,7 +466,12 @@ namespace boost
     {
         void add_thread_exit_function(thread_exit_function_base* func)
         {
-            detail::thread_data_base* const current_thread_data(get_current_thread_data());
+            detail::thread_data_base* current_thread_data(get_current_thread_data());
+            if(!current_thread_data)
+            {
+                make_external_thread_data();
+                current_thread_data=get_current_thread_data();
+            }
             thread_exit_callback_node* const new_node=
                 heap_new<thread_exit_callback_node>(func,
                                                     current_thread_data->thread_exit_callbacks);
@@ -523,14 +505,14 @@ namespace boost
             return NULL;
         }
         
-        void set_tss_data(void const* key,tss_cleanup_function func,void* tss_data,bool cleanup_existing)
+        void set_tss_data(void const* key,boost::shared_ptr<tss_cleanup_function> func,void* tss_data,bool cleanup_existing)
         {
             tss_cleanup_implemented(); // if anyone uses TSS, we need the cleanup linked in
             if(tss_data_node* const current_node=find_tss_data(key))
             {
                 if(cleanup_existing && current_node->func)
                 {
-                    (current_node->func)(current_node->key,current_node->value);
+                    (*current_node->func)(current_node->value);
                 }
                 current_node->func=func;
                 current_node->value=tss_data;
diff --git a/src/win32/tss_pe.cpp b/src/win32/tss_pe.cpp
index 3aef63a1..4528b90f 100644
--- a/src/win32/tss_pe.cpp
+++ b/src/win32/tss_pe.cpp
@@ -120,6 +120,16 @@ extern "C" const IMAGE_TLS_DIRECTORY32 _tls_used __attribute__ ((section(".rdata
         //The .CRT$Xxx information is taken from Codeguru:
         //http://www.codeguru.com/Cpp/misc/misc/threadsprocesses/article.php/c6945__2/
 
+#if (_MSC_VER >= 1400)
+#pragma section(".CRT$XIU",long,read)
+#pragma section(".CRT$XCU",long,read)
+#pragma section(".CRT$XTU",long,read)
+#pragma section(".CRT$XLC",long,read)
+        static __declspec(allocate(".CRT$XLC")) _TLSCB __xl_ca=on_tls_callback;
+        static __declspec(allocate(".CRT$XIU"))_PVFV p_tls_prepare = on_tls_prepare;
+        static __declspec(allocate(".CRT$XCU"))_PVFV p_process_init = on_process_init;
+        static __declspec(allocate(".CRT$XTU"))_PVFV p_process_term = on_process_term;
+#else
         #if (_MSC_VER >= 1300) // 1300 == VC++ 7.0
         #   pragma data_seg(push, old_seg)
         #endif
@@ -144,7 +154,6 @@ extern "C" const IMAGE_TLS_DIRECTORY32 _tls_used __attribute__ ((section(".rdata
             #pragma data_seg(".CRT$XLB")
             _TLSCB p_thread_callback = on_tls_callback;
             #pragma data_seg()
-
             //Callback for termination.
 
             #pragma data_seg(".CRT$XTU")
@@ -153,6 +162,7 @@ extern "C" const IMAGE_TLS_DIRECTORY32 _tls_used __attribute__ ((section(".rdata
         #if (_MSC_VER >= 1300) // 1300 == VC++ 7.0
         #   pragma data_seg(pop, old_seg)
         #endif
+#endif
 
         PVAPI on_tls_prepare(void)
         {