I usually iterate between code and UML diagrams. I find that the diagrams help to show the big picture and a solid path forward and they can be changed rather quickly when issues are discovered. Also, when I have the diagrams, the coding tends to become trivial.

Conversely, when I draw the code in pictures it exposes dependency issues and makes design improvement opportunities blatantly obvious, which probably wouldn't have been noticed if we just had the code to work from. In particular, if it is a pain to draw, then it is going to also be a pain to code and a nightmare to maintain.

I have found that iteration is required, because just drawing the pictures always misses important aspects, while just coding results in problematic designs.

However, like another poster said, it all boils down to the people. If they are skilled "using" UML diagrams then UML is invaluable. If they aren't then the diagrams are of no value.

Thus, the answer to your question really is, "it depends". In this case, it depends on the people, the complexity of the project and how important it is that the system works properly.