I can't comment on the METT practice test because I actually did the long-version of the training program last year. For what it's worth, I managed to earn 100% at the fastest speed after the training program. I can't remember how I did on the initial practice test, but I know it wasn't 100%. The training course itself is worth the price, I think, especially if you can afford the longest version of the course available (it has more practice runs before the final test). I liked it because it actually said how each expression was constructed and what the essential elements were; if any of those essential elements were missing, it was likely a fake expression. It gives you the opportunity to practice at slower speeds and lets you increase the speed gradually.
The METT was developed by Paul Ekman, who's been studying the connection between emotions and facial expressions for 40+ years. Do a search for his books; you can also get them from the METT site. He goes more in-depth into what the facial expressions mean. I have one of his books that I use as a reference because it really breaks down the different components of the facial expression and what the underlying emotion is.
If you want to really get into the subject, look up the Facial Action Coding System. Ekman was one of the researchers who helped develop it. It divides the face into different regions based on the underlying musculature and shows, from a more "mechancial" perspective, how emotions are expressed. From what I understand the system is popular with some artists and animators because it's supposed to help them create more realistic facial expressions.
As other posters have said, it's often easier for me to recognize expressions, and body language in general, when I'm not actually engaged in conversation with someone. Although I remember one time I went to a restaurant alone and the hostess asked if i wanted to sit at the counter. I asked for a booth and I caught her showing the micro-expression for disgust. I had to stifle a laugh as she seated me.