Note: written before parent node was updated.

Good idea, but far less than enough hubris. You can replace the entire repetitive beginning of the script by just a few lines:

sub make_rx_from {
    open my $fh, "<", shift or warn("Can't open fromgood $!\n"), retur
+n;
    chomp(my @line = <$fh>);
    return map qr/$_/, @line;
}

my @togood       = make_rx_from "togood";
my @receivedgood = make_rx_from "receivedgood";
my @badcontent   = make_rx_from "contentbad";
my @maillists    = make_rx_from "maillist";
my @badwords     = make_rx_from "badwords";
my @frombad      = make_rx_from "frombad";
my @fromgood     = make_rx_from "fromgood";
[download]

But that's still far from ideal. Look at your code's main loop: you are examining a specific header, testing it against a whitelist and a blacklist, flagging accordingly. The same pattern every time. That can be factored out along these lines:

HEADER: foreach my $line (@$messref) {
    # detects first blank line go do body check of first 10 body lines
    last if $line =~ /^\s*$/;

    my ($header) = grep $line =~ /^$_:/i, keys %test_for;
    next unless $header;

    for my $test (qw(ok bad)) {
        my @match = grep $line =~ $_, @{ $test_for{$header}{$test} };
        next if not @match;
        $flag = $test; 
        print "Message $test - header $header matched: @match\n";
        last HEADER;
    }
}
[download]

You could also check the message size prior to any other tests for efficiency. I'll post the complete code as I refactored it in a separate node, for easier downloading.

Makeshifts last the longest.