It depends on what you want to do with the ADC values, and how fast you need to do it. If you need to take some time-critical action based on the ADC values, you should deal with that in the interrupt. However, if this is the case, to save yourself the expense of first calculating the "real" analog value from the ADC counts, you should just express your critical thresholds in terms of the ADC counts.
For example, if your ADC outputs 8 counts per volt, and you need to fire an emergency stop if the ADC reports greater than 1.5 volts, you write your interrupt code to fire the E-stop at 12 counts. You could then still pass the ADC counts off to the main loop for translation into user-friendly units for user interface purposes, etc.
The general principle is that to guarantee you can meet your deadlines, you structure your application and any settings to make it as easy as possible to meet those deadlines. If something doesn't have a hard deadline, put it in the main loop (again, in general).