If you want to use the 95% of a 20 minute test rather than testing it with a full, flat out one hour effort, then you should follow the protocol from which that approximation was derived. IIRC, you should warm up, then do 3x1 minute high cadence efforts, then rest, then do a maximal 5 minute effort, rest for 10 mins, then do a 20 minute maximal effort and cool down. You then approximate FTP from the 20 minute effort. Without the 5 minute maximal effort, you will probably over estimate your FTP, then when you try to pace to it later, in a real situation like a 25 mile time trial, you will blow up catastrophically.
You should test it in the context which you usually train. I historically tested mine on the turbo because I trained on the turbo exclusively, if you train mostly on the road, test on the road, train mostly on the turbo, test on the turbo. Just be aware there will most likely be a difference, in most cases, with power being higher on the road than the turbo. I have now managed to close the gap between turbo and road power and have split my training up with more on the road, so I just take my power from a 25 mile time trial.
There are other ways to estimate FTP too, such as from a MAP (maximal aerobic power) test, considering the data my partner got from a map test and what she is able to product on the road, the MAP test over estimated her FTP by a large amount. I've not done a MAP test but it basically involves increasing power every minute, until you blow, I do not see the point in blowing that hard if there is no prize money or pride at stake.