David: Thanks for your reply!

To answer one of your questions, "Would you look for clients by their responses across survey's?" - At the moment, I don't see that happening for particular questions, but you never know. At the moment, the main purpose of the survey would be to define a total risk level for the client. If the client's risk level is say "high", then the investigator/user would know to do certain actions to help the client. I also imagine the users wanting annual summary stats such as: How many clients last year had a risk level of low? med? high?

You do not have to explain, but I thought I would let you know that I didn't understand this part of your reply: "But if Question,Person and survey were normalized types out of some entity table, and person-has-a-survey, survey-has-a-question and Person-takes-a-survey-question where normalized types out of some entity-relationship table then I would start to squirm."

As for this part: "If you know that the client takes a survey not a supervisor or an operator...", I'm not sure I understood your full point, but I can comment on this one part. The clients themselves will never actually be filling out the "survey". Instead, the operator (our investigators/staff) will decide if a survey/assessment is required. If so, then a completely survey will likely be set up for that client, by which I mean a stored proc will likely be called to generate a record for every active question and attach to that client. The default answer for each question would be null or "not applicable", and the user would just change the answers for the questions that apply. For now, I think that set-up would allow me to create the most user-friendly interface.

Thanks again for your interest.