This article is from the Training Your Dog FAQ, posted to rec.pets.dogs newsgroup. Maintained by Cindy Tittle Moore with numerous contributions by others.
B.F. Skinner outlined the principles of what he termed "operant conditioning." In contrast to classical conditioning, in operant conditioning the reinforcement cycle starts with some action on the part of the trainee (in Skinner's language, the operant). Operant conditioning is therefore _always_ dependent on behavior, whereas classical conditioning is _not_. We have:
1. dog does something (operant behavior)
2. dog gets food (positive reinforcement)
Under this theory, if we control which behaviors are reinforced, we should be able to get the dog to offer those behaviors more often. If the dog gets good stuff in association with a particular behavior, he's likely to repeat it; if something bad happens, he's less likely to repeat it. In practical training terms, this means that if Andy picks up his dumbbell (step 1), Andy gets some turkey (step 2); if he doesn't, he doesn't get the turkey. The result should be that in the long run, Andy will grab the dumbbell eagerly, even if he isn't a natural retriever.