This doesn't change your functionality at all (i don't think), but does shorten all the code up and hopefully makes it much easier to modify from here.. Basically it makes use of hashes instead of having to declare a bunch of variables (e.g. $a, $g, $t, $c) so that everything is easier to work with.

One thing i didn't understand was the $w usage -- i assumed in my code that you want $w to be the length of the longest item, but your code was setting it to be the length of the last item...

use strict;
use warnings;

my @instances  = qw ( AAAAA ATCGA ATAAA );
my @instances2 = qw ( AAAAA AACGA ATAAA ATAAA );

print consensus(@instances),"\n";        # ATAAA
print consensus(@instances2),"\n";       # ATAAA
exit;

sub consensus{
 my @mi = @_;
 chomp(@mi);
 my $motif_count=0;
 my @words =();

  my %H = ( A=>[], T=>[], C=>[], G=>[] );

  s/\s//g for @mi;
  my ($w) = sort {$b <=> $a} map {length} @mi;    # set w to the lengt
+h of the longest element

    foreach my $j ( 0 .. $w-1 ){
        # Initialize the base counts.
        my %h = ( a=>0, t=>0, c=>0, g=>0 );
        foreach my $mi ( @mi ) {
            $h{ lc substr($mi,$j,1) }++;           # example:  $h{g}++
+;
        }
        push @{$H{ uc $_ }}, $h{$_} for keys %h;   # example:  push @{
+$H{G}}, $g;
    }

    my @cons = ();
    my %prefOrder = ( A=>1, T=>2, C=>3, G=>4 );
    foreach my $B ( 0 .. $w-1 ){
      push @cons, [ sort { $H{$b}->[$B] <=> $H{$a}->[$B] || $prefOrder
+{$b} <=> $prefOrder{$a} } qw/A T G C/ ]->[0];
    }

    return @cons;
}
[download]

Update: more efficient way to set %H would be:

  my %H = ( A=>[], T=>[], C=>[], G=>[] );
  my @mi_letters = map { [split '', uc $_] } @mi;
  foreach my $j ( 0 .. $w-1 ){
    $H{ $_->[$j] }->[$j]++ for @mi_letters;
  }

  # note: to avoid warnings, change the sort further down to include t
+his (the '||0'):
  #    ($H{$b}->[$B]||0) <=> ($H{$a}->[$B]||0)
[download]