diff --git a/memcheck/docs/mc-manual.xml b/memcheck/docs/mc-manual.xml
index ea1d36363..0a0a0780d 100644
--- a/memcheck/docs/mc-manual.xml
+++ b/memcheck/docs/mc-manual.xml
@@ -46,9 +46,6 @@ Memcheck can detect the following problems:</para>
     <computeroutput>memcpy()</computeroutput> and related
     functions</para>
   </listitem>
-  <listitem>
-    <para>Some misuses of the POSIX pthreads API</para>
-  </listitem>
 </itemizedlist>
 
 </sect1>
@@ -159,19 +156,25 @@ Memcheck can detect the following problems:</para>
 
   <listitem id="partial">
     <para><computeroutput>--partial-loads-ok=yes</computeroutput>
-    [default]</para>
-    <para><computeroutput>--partial-loads-ok=no</computeroutput></para>
-    <para>Controls how Memcheck handles word (4-byte) loads from
+    </para>
+    <para><computeroutput>--partial-loads-ok=no</computeroutput>[default]
+    </para>
+    <para>Controls how Memcheck handles word-sized, word-aligned loads from
     addresses for which some bytes are addressible and others are
-    not.  When <computeroutput>yes</computeroutput> (the
-    default), such loads do not elicit an address error.
+    not.  When <computeroutput>yes</computeroutput>,
+    such loads do not elicit an address error.
     Instead, the loaded V bytes corresponding to the illegal
-    addresses indicate undefined, and those corresponding to
+    addresses indicate Undefined, and those corresponding to
     legal addresses are loaded from shadow memory, as usual.</para>
-    <para>When <computeroutput>no</computeroutput>, loads from
-    partially invalid addresses are treated the same as loads
-    from completely invalid addresses: an illegal-address error
+    <para>When <computeroutput>no</computeroutput>(the default), 
+    loads from partially invalid addresses are treated the same as 
+    loads from completely invalid addresses: an illegal-address error
     is issued, and the resulting V bytes indicate valid data.</para>
+    <para>Note that code that behaves in this way is in violation of
+    the the ISO C/C++ standards, and should be considered broken.
+    If at all possible, such code should be fixed.  This flag should
+    be used only as a last resort.
+    </para>
   </listitem>
 
   <listitem id="strlen">
diff --git a/memcheck/mac_shared.c b/memcheck/mac_shared.c
index fb6b01a6c..bd303be2d 100644
--- a/memcheck/mac_shared.c
+++ b/memcheck/mac_shared.c
@@ -58,7 +58,7 @@
 /*--- Command line options                                 ---*/
 /*------------------------------------------------------------*/
 
-Bool          MAC_(clo_partial_loads_ok)       = True;
+Bool          MAC_(clo_partial_loads_ok)       = False;
 Int           MAC_(clo_freelist_vol)           = 5000000;
 LeakCheckMode MAC_(clo_leak_check)             = LC_Summary;
 VgRes         MAC_(clo_leak_resolution)        = Vg_LowRes;
@@ -100,7 +100,7 @@ void MAC_(print_common_usage)(void)
 "    --leak-check=no|summary|full     search for memory leaks at exit?  [summary]\n"
 "    --leak-resolution=low|med|high   how much bt merging in leak check [low]\n"
 "    --show-reachable=no|yes          show reachable blocks in leak check? [no]\n"
-"    --partial-loads-ok=no|yes        too hard to explain here; see manual [yes]\n"
+"    --partial-loads-ok=no|yes        too hard to explain here; see manual [no]\n"
 "    --freelist-vol=<number>          volume of freed blocks queue [5000000]\n"
 "    --workaround-gcc296-bugs=no|yes  self explanatory [no]\n"
    );
diff --git a/memcheck/mc_main.c b/memcheck/mc_main.c
index 96ce84d00..ba8183960 100644
--- a/memcheck/mc_main.c
+++ b/memcheck/mc_main.c
@@ -417,7 +417,7 @@ ULong mc_LOADVn_slow ( Addr a, SizeT szB, Bool bigendian )
    SizeT i           = szB-1;
    SizeT n_addrs_bad = 0;
    Addr  ai;
-   Bool  aok;
+   Bool  aok, partial_load_exemption_applies;
    UWord abit, vbyte;
 
    PROF_EVENT(30, "mc_LOADVn_slow");
@@ -436,7 +436,25 @@ ULong mc_LOADVn_slow ( Addr a, SizeT szB, Bool bigendian )
       i--;
    }
 
-   if (n_addrs_bad > 0)
+   /* This is a hack which avoids producing errors for code which
+      insists in stepping along byte strings in aligned word-sized
+      chunks, and there is a partially defined word at the end.  (eg,
+      optimised strlen).  Such code is basically broken at least WRT
+      semantics of ANSI C, but sometimes users don't have the option
+      to fix it, and so this option is provided.  Note it is now
+      defaulted to not-engaged.
+
+      A load from a partially-addressible place is allowed if:
+      - the command-line flag is set
+      - it's a word-sized, word-aligned load
+      - at least one of the addresses in the word *is* valid
+   */
+   partial_load_exemption_applies
+      = MAC_(clo_partial_loads_ok) && szB == VG_WORDSIZE 
+                                   && VG_IS_WORD_ALIGNED(a) 
+                                   && n_addrs_bad < VG_WORDSIZE;
+
+   if (n_addrs_bad > 0 && !partial_load_exemption_applies)
       MAC_(record_address_error)( VG_(get_running_tid)(), a, szB, False );
 
    return vw;