In the Groundwork of the Metaphysics of Morals Kant presents a hard and fast standard for how to determine whether something is a duty. There are duties which apply in all cases, "perfect duties," like for example never lie. Then there are duties which one should follow unswervingly, but one can choose when to apply, "imperfect duties" like cultivating one's talents. The measure of a perfect duty is that when you make it into a universal law, it leads to a conceptual inconsistency, leading you to conclude that you should never do it. The measure of an imperfect duty is that if you make it into a universal it would somehow lead to an empirical inconsistency, in that it violates your natural instincts. I'll explain the empirical inconsistency first, since it's a bit more unclear. We'll use his example, cultivating your talents (4:423). Kant basically says that you couldn't possibly spend your whole life not cultivating your talents since it would violate your natural instincts. It seems like, if you were to try to do this, something deep inside of you would scream that it's wrong or that it's undesirable or substandard for a human or something like that. Though I think Kant has greater confidence in the rationality of our usual moral instincts and he and I probably would disagree on the details of what things are imperfect duties, I nonetheless think this seems like a sound idea. There're certain habits that just seem less admirable or worthy of respect and we should try to do their opposite when we can--like cultivate our talents, help others, share, learn, create, etc.
The one that seems to me most crazy is the perfect duty and the conceptual inconsistency. The example he uses is the lying promise (4:422). If I'm tempted to make a promise I know I can't fulfill (like borrow money I can't pay back) should I make such a promise? Kant says that we have a perfect duty not to make a lying promise because if we were to make it a universal law that all people should always make lying promises, then no one would be able to promise anything because people would know it always to be a lie. We would have a conceptual inconsistency of lie and promise, which shows us that one should never make a lying promise.
The first limitation one should notice, is perfect duties can only be "thou shalt nots." The conceptual inconsistency is the measure of a perfect duty, but it only shows when something should never be done. The next thing is that the conceptual inconsistency doesn't work well for many things that we would assume to be perfect duties, like don't kill for example. If I were to make it a universal maxim that I should kill anyone I don't like, then there would be a lot of people getting killed. But where's the conceptual inconsistency? Also, why do you even need to universalize it to see the inconsistency? Isn't a lying promise already inconsistent? The other thing is that, this rubric of determining what is a perfect duty is based on the seeing the consequences of universalizing a rule. But consequences are uncertain. This is one of the reasons why Kant strays away from consequentialism, why the morality of an act is based on the act itself, not its consequences. Also, the consequences that Kant envisions as the result of universalizing these maxims seem rather superficial and static. If we were to universalize the rule that people should make lying promises whenever it is convenient, then wouldn't people try to find ways to make contracts without having to rely on people's unreliable promises? Sounds something like the world we live in, doesn't it? Don't people already avoid breaking many promises for selfish reasons? Do people keep promises with their friends out of duty, or because they don't like to take advantage of their friends and don't want to lose their friends? Do people avoid violating business contracts out of duty, or because they want to make future contracts with the same people or because they're worried about the legal repercussions of violating a contract? We already know that promises can be unreliable, so we invent techniques and create institutions to force people to keep their promises. The problem is far more often reneging on promises rather than making them falsely to begin with. People are creative and dynamic, and Kant's consequences of universalizing maxim assume people are static and simple.
The ground of the perfect duties seems quite flimsy. Maybe one could accept this imperfect duty, and the almost inevitable subjectivity of it, but to think that reason gives us access to universal rules is foolish.