diff --git a/lib/utf8/all.pm b/lib/utf8/all.pm
index 5574c4a..cdeaf93 100644
--- a/lib/utf8/all.pm
+++ b/lib/utf8/all.pm
@@ -110,15 +110,21 @@ L<Cwd::utf8> for fully utf-8 aware Cwd functions.
 use Import::Into;
 use parent qw(Encode charnames utf8 open warnings feature);
 use Symbol qw(qualify_to_ref);
+use Config;
 
 # Holds the pointers to the original version of redefined functions
 state %_orig_functions;
 
 sub import {
+    my $class = shift;
+    
     # Enable features/pragmas in calling package
     my $target = caller;
+
+    my $utf8_encoding = $class->_choose_utf8_encoding;
+    
     'utf8'->import::into($target);
-    'open'->import::into($target, qw{:encoding(UTF-8) :std});
+    'open'->import::into($target => $utf8_encoding, ':std');
     'charnames'->import::into($target, qw{:full :short});
     'warnings'->import::into($target, qw{FATAL utf8});
     'feature'->import::into($target, qw{unicode_strings}) if $^V >= v5.11.0;
@@ -208,6 +214,31 @@ sub _utf8_glob {
     }
 }
 
+sub _choose_utf8_encoding {
+    # No threads? No problem.
+    return ':encoding(UTF-8)'    if !$Config{usethreads} && !$Config{useithreads};
+
+    # 5.24.0 seems to have fixed the major utf8 issues.
+    return ':encoding(UTF-8)'    if $^V ge 5.24.0;
+
+    # A safe default.
+    return ':utf8';
+}
+
+=head1 WHICH UTF-8 ENCODING?
+
+I<TL;DR>. Perl's unicode has bugs. utf8::all will try to work around them.
+
+As of this writing, Perl has several ways to do utf-8. It has to do
+with whether "unassigned" code points are considered errors or
+not. The details are in L<perlunicode: Noncharacter code points|http://perldoc.perl.org/perlunicode.html#Noncharacter-code-points>. Perl also has lots of Unicode bugs, particularly with threads and
+strict utf-8 encoding (ie. C<:encoding(UTF-8)>).
+
+utf8::all will prefer the strictest encoding available, but it may
+choose a less strict utf-8 encoding if it detects your Perl is
+vulnerable to Unicode bugs. This should have no effect on how valid
+utf-8 is handled.
+
 =head1 INTERACTION WITH AUTODIE
 
 If you use L<autodie>, which is a great idea, you need to use at least version
diff --git a/t/lexical-again.t b/t/lexical-again.t
index a1f4a05..6f2877a 100644
--- a/t/lexical-again.t
+++ b/t/lexical-again.t
@@ -2,7 +2,7 @@
 # no utf8::all should disable its effects lexically
 # Note: Changes to @ARGV, STDIN, STDOU, and STDERR are always global!
 
-use Test::More tests => 17;
+use Test::More;
 use PerlIO;
 
 my $expected_unicode = "\x{30c6}\x{30b9}\x{30c8}"; # Unicode characters
@@ -52,7 +52,6 @@ for my $fh (keys %handles) {
     my @layers = PerlIO::get_layers($handles{$fh});
     ok(grep(m/utf8/, @layers), "$fh: utf8 does appear in the perlio layers")
         or diag explain { $fh => \@layers };
-    ok(grep(m/utf-8-strict/, @layers), "$fh: utf-8-strict does appear in the perlio layers")
-        or diag explain { $fh => \@layers };
 }
 
+done_testing;
diff --git a/t/open.t b/t/open.t
index b782395..227fa8d 100644
--- a/t/open.t
+++ b/t/open.t
@@ -2,14 +2,14 @@
 # Test opening an actual file
 use utf8::all;
 use PerlIO;
-use Test::More tests => 4;
+use Test::More;
 
 ok open my $in, '<', 'corpus/testfile';
 my @layers = PerlIO::get_layers($in);
 ok(grep(m/utf8/, @layers), 'utf8 appears in the perlio layers')
     or diag explain { $fh => \@layers };
-ok(grep(m/utf-8-strict/, @layers), 'utf-8-strict appears in the perlio layers')
-    or diag explain { $fh => \@layers };
 
 my $contents = do { local $/; <$in>};
 is $contents, "\x{30c6}\x{30b9}\x{30c8}\n", 'unicode retrieved OK';
+
+done_testing;
diff --git a/t/threads.t b/t/threads.t
new file mode 100644
index 0000000..4ff1d27
--- /dev/null
+++ b/t/threads.t
@@ -0,0 +1,33 @@
+#!perl
+
+# Test that utf8::all is choosing the right encoding to not
+# tickle thread bugs.
+
+use strict;
+use warnings;
+
+# This is loaded before threads. It will not be aware of tests run in
+# a thread.
+use Test::More 0.96;
+use Config;
+
+BEGIN {
+    plan skip_all => "Requires threads"
+      if !$Config{usethreads};
+}
+
+# Deliberately before loading threads so we don't cheat and check
+# if threads are loaded, that would be brittle.
+use utf8::all;
+
+use threads;
+use threads::shared;
+
+note "basic utf8 + threads bug"; {
+    my $ok :shared = 0;
+    my $t = threads->create(sub { $ok = 1; });
+    $t->join();
+    ok $ok, "threads ok with utf8::all";
+}
+
+done_testing;
diff --git a/t/utf8.t b/t/utf8.t
index b767890..fba32b2 100644
--- a/t/utf8.t
+++ b/t/utf8.t
@@ -19,8 +19,6 @@ use Test::More;
         my @layers = PerlIO::get_layers($fh);
         ok(grep(m/utf8/, @layers), 'utf8 appears in the perlio layers')
             or diag explain { $fh => \@layers };
-        ok(grep(m/utf-8-strict/, @layers), 'utf-8-strict appears in the perlio layers')
-            or diag explain { $fh => \@layers };
     }
 }
 
@@ -33,7 +31,7 @@ use Test::More;
     END { unlink "perlio_test2" }
 
     my @layers = PerlIO::get_layers($test_fh);
-  SKIP: {
+    SKIP: {
         # If we have the Perl Unicode flag set that adds the UTF-8 layer,
         # we need to skip this test.
         skip 'Perl Unicode flag set that always adds UTF-8 layer to output', 1 if (${^UNICODE} & 16);