A better tr

private static readonly HashSet<char> badChars =
    new HashSet<char> { '!', '@', '#', '$', '%', '_' };

public static string CleanString(this string str)
{
    var result = new StringBuilder(str.Length);
    for (int i = 0; i < str.Length; i++)
    {
        if (!badChars.Contains(str[i]))
            result.Append(str[i]);
    }
    return result.ToString();
}

This algorithm makes use of the .NET 3.5 'HashSet' class to give O(1) look up time for detecting a bad char. This makes the overall algorithm O(n) rather than the O(nm) of your posted one (m being the number of bad chars); it also is lot a better with memory usage, as explained above.

>>5
That's but a small part of my aversion to tedious explanation. AFAIK to the average /prog/lodyte, >>2 might as well be using a Y-combinator--which I suspect is better understood around here... and at least it has an evocative name and possibly bewildering implementation to wield against people who would write it off.