This is an excellent question. The difference between Classical conditioning (also called Pavlovian conditioning) and operant (instrumental) conditioning is subtle for the new student, but can be quite profound when fully appreciated.
Pavlovian conditioning is learning a response that you have no control over. In this context, a conditioned taste aversion (CTA) might be produced by mildly poisoning a rat after it eats watermelon for the first time. Or you might suffer food poisoning after eating a watermelon. The CS is the watermelon. The usual response to watermelon is licking lips and paws, and savouring the sweet flavour - any rat version of yum you can think of. However after poisoning (where the US is usually denoted by the poisoning method, e.g. weak lithium chloride injections), the new response is gaping, retching and avoiding the now yucky flavour. In people who get food poisoning, we know the actual flavour of the food changes from pleasant to revolting, and can even elicit retching and vomiting. This response won't change even if you tell the person the next watermelon is fresh and sweet. In other words, the subject of the CTA cannot change their response (yucky flavour/retching) to the CS at will. This obviously creates a problem when a CTA is produced as a side-effect of medical treatment such as chemotherapy. Which is why people on chemo will often only be fed very bland and boring foods (to reduce the chance of a CTA). Even telling the person the chemo is causing the nausea is not sufficient to change their opinion of the foods they ate shortly before they felt sick.
By contrast, instrumental conditioning is learning a response you have control over. E.g., training a rat to press a lever for watermelon (or a person buying watermelon from the grocer). In this case you or the rat has control over the action (lever pressing or shopping at the grocer). So after a rat is convinced to press a lever for sweet juicy watermelon, if the rat gets poisoned by the watermelon then it will simply stop pressing the lever. Here the instrumental response to a CTA is 'avoidance'. Just as you would stop shopping at a grocer who previously sold you a bad watermelon. If however the owner of the grocery changed, you might be happy to resume shopping for watermelon there. Note that while you may no longer avoid shopping for watermelon, your Pavlovian response to the bad watermelon might still be intact - i.e., you might retch when you first bite into the fresh watermelon you just bought. This can obviously create odd situations where animals will work for food they do not want.
Whenever experimental psychologists are faced with a learned response they don't understand, one of the first things they will do is ask whether the response is Pavlovian or instrumental. The classic way to test this is with an omission schedule. Under an omission schedule, the reward is removed if the animal makes the response. If the animal continues to make the response, then the response is clearly not under instrumental control. If you look up some of Anthony Dickinson's work on this subject, you will discover many weird and wondrous examples of this distinction, especially in his incentive learning research. A lot of behaviour we think of as instrumental, is probably Pavlovian - e.g., a dog approaching a food bowl when he hears the can opener. One place to start is Dickinson, 1981, Conditioning and associative learning, British Medical Journal, 37 (2), 165-168