note ted.byers OK. Slapping ideas together, and extracting commonalities, from among the responses so far and combining with constructing a random string, I tried the following experiment. <code> use strict; use warnings; use Math::Random::MT::Auto::Range; binmode(STDOUT, ":utf8"); $| = 1; #the following exhausts all memory #my $all = join q[], grep /^\w$/, map chr, shuffle 0 .. 0x10FFFF; #print $all,"\n"; # # so I tried: my @chars = map(chr, 0 .. 0x10FFFF); my $rng = Math::Random::MT::Auto::Range->new(LO => 0, HI => 0x10FFFF, TYPE => 'INTEGER'); # three options for doing the basically the same thing for (my $i = 0 ; $i < 10 ; $i++) { my $code = $rng->rrand; print "$i <=> $code <=>",chr($code),"\n"; print "\t$i <=> $code <=>",$chars[$code],"\n"; print "\t$i <=> $code <=>",pack("U",$code),"\n"; } </code>There are a couple problems with this. chr generates some errors and warnings. There are a couple thousand instances of "UTF-16 surrogate 0xd81b at c:/Work/test.utf8.latin1.pl line 16.", and over a hundred instances of "Unicode non-character 0xfdd0 is illegal for interchange at c:/Work/test.utf8.latin1.pl line 16.", each for a different integer. Line 16 is the line where @chars is initialized. I suppose a couple thousand problem characters in @chars is not a huge issue, but given that I want to use it to test functions for converting from utf-8 to latin1 and back, I expect that if they happen to occur in my sample, there'd be some false indications of errors in these functions (from the Perl package 'Encode').The second problem is that though I put the statement "binmode(STDOUT, ":utf8");" at the start of the script, the printout contains only rectangles and square where the UTF-8 character ought to be; at least when I execute within Emacs. When I execute the script in the Windows commandline terminal, I invariably get gibberish (different for each character) the width of four characters. How, then, do I actually see the characters? I thought, probably mistakenly, I'd see a few greek or sanskrit characters or characters from other alphabets.What I was thinking of doing is amend the loop I show above to construct a single string from the ten characters produced, and then use the functions from 'Encode' to convert it to latin1, and then back to utf-8, to see if the input string and the output string are the same (and if not, then this whole idea is doomed to fail). I'd repeat this test for a few million random utf-8 strings, and if there are no failures, then I could use this idea as a temporary measure until I can test our systems to see how best to adapt to use of utf-8 throughout.What, then, do I do to exclude the integers that either result in a utf-16 surrogate and those that represent illegal characters? And is it possible for me to actually see the characters produced?ThanksTed 1009778 1009778