This is one of the many qustions that the field of
Cognitive Science tries to answer.
The first thing you have to do is define 'thought'. Good luck, because that's really the tricky part.
As a CogSci major myself, I've found out enough to know that things are not cut-and-dry either way. My own idea is that certain properties of 'thinking', such as sensations, beliefs, logical chains, causal occurences, are all multiply realizable. That means, though they may have one physical form in one being (John), they might have rather different forms in a different being (Sally). And they can still be the same thing. That is, even though detection of causal occurence might be indicated by certain neural patterns in John's head, those patterns might be different in Sally's head.
This idea stretches over everything. Most obviously, it can be applied to computers. What is pain? Is there something that serves the prupose of pain in a computer? To answer this, you have to define 'pain' as abstractly as possible. That's a trick, too.
Can my computer think? I doubt it very much. Can a Cray computer think? Probably not. How about a Cray whose been programmed in a very specific way? Uh.... Or how about a hypothetical cybernetic neural network? Can that think?
We will see thinking machines in our lifetime. I'll bet dollars to donuts the idea won't even shock you in, say 15 years. We'll probably see it within 20. And those numbers are probably pessimistic.
Scientific knowledge is going to increase at a blistering pace over the next few decades, and Cognitive Science will go even faster than most. It is a very wide field, from psychology, biology, and neurology to computer science, physics, and linguistics. An advance in any of these fields will yield startling progress in CogSci. Thinking machines are coming, and soon.
Should be interesting.