How is this better than unix "cut"? I know it's different: "cut" uses 1-based column numbers instead of 0-based, and column selection requires a "-f" option flag, and when "-f" is not provided, it exits with usage instructions, instead of listing the fields on the first tab-delimited line of input. But to make it better:

• you might consider allowing options for selecting a delimiter other than tab (like "cut" does) -- e.g. the delimiter could be a regex (which is something "cut" can't do)
• you might also consider allowing the output field delimiter to be different from the input field delimiter (something else "cut" can't do)

I notice that you can output fields in arbitrary orders, and even output a given column more than once, and these are handy improvements over cut. But why stop there?

Minor nit-picks:

• don't use "chop" ("chomp" is better, but see below)
• given that it's only useful with piped input, die with usage instructions when STDIN is a tty (read about the "-t" function in "perldoc -f -X")
• it might also be nice to die with instructions when @ARGV contains things that aren't digit strings
• I like POD. Don't you?
• given a set of digit strings in @ARGV, your process could be expressed in less code:

  while (<STDIN>) {
      tr/\r\n//d;  # even better than chomp!
      print join( "\t", ( split /\t/ )[@ARGV] ), "\n";
  }
  [download]

• (added as an update) BTW, this construct: until ($line=~/\t/) { $line=<STDIN>; } will be an infinite loop if the piped input never contains a tab character. I think you'll want this instead:

  while(<STDIN>) {
      last if /\t/;
  }
  die "No tab-delimited fields found\n" unless ( /\t/ );
  [download]