>I prefer to do all validation when the user tries to save data. Some advantages:
>

The user may thus input data in any order. If you use a valid() event or something similar, the user may not even press a "Cancel" button - the error message will still appear.
>

Sometimes some validation involves more than one field.
>

It is also simpler to program that way.

For me the biggest advantage of this pattern is the ease of translation into a browser based app.
I follow it also, but to a lesser degree: I will at least validate "simple" vaildations like between (userinput, low, high) at the control level if they are constants and not dependent on other values. My rule of thumb is here that everything testable without extensive scripting in a browser app can be ported as well without the need to re-enter the process of discussing project specs. That includes nowadays lookup values - dynamically switching vales for lookups in HTML was the best feature of XmlHttp.

regards

thomas