Depending on your platform, you may use /dev/urandom or CAPICOM. This is nicely summarized in this comment from Mark Seecof:
"If you need some pseudorandom bits for security or cryptographic purposes (e.g.g., random IV for block cipher, random salt for password hash) mt_rand() is a poor source. On most Unix/Linux and/or MS-Windows platforms you can get a better grade of pseudorandom bits from the OS or system library, like this:
<?php
// get 128 pseudorandom bits in a string of 16 bytes
$pr_bits = '';
// Unix/Linux platform?
$fp = @fopen('/dev/urandom','rb');
if ($fp !== FALSE) {
$pr_bits .= @fread($fp,16);
@fclose($fp);
}
// MS-Windows platform?
if (@class_exists('COM')) {
// http://msdn.microsoft.com/en-us/library/aa388176(VS.85).aspx
try {
$CAPI_Util = new COM('CAPICOM.Utilities.1');
$pr_bits .= $CAPI_Util->GetRandom(16,0);
// if we ask for binary data PHP munges it, so we
// request base64 return value. We squeeze out the
// redundancy and useless ==CRLF by hashing...
if ($pr_bits) { $pr_bits = md5($pr_bits,TRUE); }
} catch (Exception $ex) {
// echo 'Exception: ' . $ex->getMessage();
}
}
if (strlen($pr_bits) < 16) {
// do something to warn system owner that
// pseudorandom generator is missing
}
?>
NB: it is generally safe to leave both the attempt to read /dev/urandom and the attempt to access CAPICOM in your code, though each will fail silently on the other's platform. Leave them both there so your code will be more portable."