I just took the BPMN Model understanding self test. It is a research project of Humboldt University of Berlin, that I can recommend. I was asked about 30 models and how I understand them. The test takes about 30 minutes and is a nice excercise.
I made it to rank 14 of 394. However I wonder who the 13 were, that were better 
. So give it a try, maybe you can beat me. The test is anonymous however.
Nevertheless I want to share what I thought when I saw the models. They were quite complicated. I think if models are as complicated in a real project as those in the test, then something went terribly wrong in the first place. I agree, that it is fine for a research project to use artifical complicated models, to find out more about human model comprehension. And I am very interested in the research result. But models must be much simpler than those.
Simpler models could be reached by limiting the scope of one model – i.e. splitting it up in different parts, using submodels for example. As far as I remember human comprehension can assess 7 items at once, not more. So in essence I think a model should not contain more than about 7 important steps.
Also it can mean to model only the most important cases and model the special cases in a different model.
And it can mean to question, if BPMN is the right model language for the purpose chosen. I know that BPMN is popular and becomes even more, because it is a standard. But in my eyes the question remains, if the task flow oriented modeling it does is really the best way to do it. In my eyes it should be evaluated as a result of this research project, if goal driven and constraint based modeling would not result in much easier models.
