Just constant practice is the key. Get used to the fact that the sheer majority of your projects will never be finished and move on - the most important thing is to enjoy yourself and become a better mapper. But still, know when you should stick by a project and try to follow through to the end. Mod-work is good since it forces you into commitment at times.

Work process varies for levels, from mapper to mapper. Some people meticulously plan every little facet of the level, while others just open Hammer and throw down some brushes. It just depends on how you work. I know ReNo, for example, usually makes 1 detailed concept room to see if he likes the style, then will continue from there.

It also helps to accept the fact that you aren't 100% correct about your own maps, accept criticism from others, and act upon it. You can't get too enamored with your own work and let personal pride get in the way. I've released two maps, Torque and Cataclysm, and in the first one I let myself become content with what I had and unwilling to change anything. I let a golden opportunity pass with that one, as it had the gameplay, but not the looks. For cataclysm when I was close to thinking it was finished people brought up the texturing and some other issues, and since I had learned from Torque I took it all to heart and totally retextured the map. It became far better because I forced myself to act on the criticism and accept that I may not know best in all situations.

I've found, mainly what separates the "bad" from the "good" from the "great" is the detail of the map. A "bad" mapper will just chuck a bunch of blocks together. It might play good but it looks ugly (but thats ok, its a learning experience). The "good" mapper will improve on that, putting in little details such as door frames, beveled edges etc. The "great" mapper takes that extra care, does all sorts of nifty things that I cant think of right now.

Above all, if you want to be a better mapper, practise. As often as you can, for as long as you can.