I like toolic's 2-D hash suggestion. However, sometimes I use just a 1-D hash. Depending upon what you are doing, this can lead to some simplifications, e.g. printing a HoH takes 2 foreach loops, a simple 1-D hash just takes one. On the other hand some operations may become more complicated. I'm just presenting another option that sometimes is a good choice. Mileage varies!

The relevant line to change n toolic's code is this one:

    #$hash{$row}{$cols[$i+1]} = $vals[$i]; #toolic's
    $hash{"$row,$cols[$i+1]"} = $vals[$i]; #1-D version
[download]

Now,

foreach (sort keys %hash)
{
   print "$_  $hash{$_}\n";
}
__END__
prints:
aaa,bar  456
aaa,baz  789
aaa,foo  123
bbb,bar  654
bbb,baz  321
bbb,foo  987
[download]

The pitfall in this method is that some character or sequence of characters that cannot be in the row or column header names should be used to join the row,col names together. You want to guarantee that running together any row,col results in a unique name. This is usually not a problem and "," is often a good choice.

Have fun!