no...do them on the road. Think about it like using different levels of power...you're power should stay relatively constant (above normal pace power) while your speed changes according to terrain, wind, etc... If you were running regularly, you'd run the hill slower than normal pace, so running tempo...you should expect to run the hill at slower than normal tempo pace (but faster than you normally run the hill)

I tend to do tempo runs based on perceived exertion...not so much a slave to pace, especially since they are often time based, not distance. That tends to take terrain out of the equation. Hope that makes sense

Personally, when I think of "tempo" runs I think of somethign similar to what you described - 4 mins faster w/ 4 mins. recovery 3 or 4 times - and I tend to do those on open roads. Speedwork, such as Yasso 800s or 1- to 2-mile intervals, I do on a track. While it is easier to maintain speed on a track (ie, no traffic lights, intersections), I couldn't imaging doing extended tempo sessions on a track. Gets boring.

Think of a tempo run as a hard run. Nothing more. The initial 15 minutes is your warm-up and should be done at a conversational pace (where you're not breathing so hard that you can't talk to someone). For the tempo part, you want to pick up the pace and run hard, but comfortable. For example, if you're a 20 min 5k runner (6:30/mi) and do your long/easy runs around 7:45-8:00, you want to do your tempo runs around 7:00. I personally just tack on 0:20-0:30/mi to my open race pace for the distance I'm doing.

The 4 min hard, 4 min recovery is what most people will call a fartlek. Depending on where in the season you are (base, build, early race, taper) I will do anywhere from none to 3 of these a week. The goal here is not to run the recovery too slowly, though. Do this by assuring you don't run too fast during the "on" parts. For the above example, I would do something like 4 miles of 1/2 mile on, 1/2 mile off in 3:15 (6:30/mi - 5k race pace), 3:45 (7:30/mi - faster than long run, but still able to recover). This gets you 2 miles at race pace, but forces you to maintain pace beyone the race distance. Also, notice that the average pace is 7:00/mi, which is right around where one should be for a tempo run of similar distance, just broken up into race effort/recovery.

As for location, you can do either workout (tempo or fartlek) whereever you feel comfortable. Bike/running paths are perfect if you have one nearby. It's painfully annoying to be running on sidewalks and stopping every 2 blocks when you're trying to keep up the pace.

I would like to find a flat course for tempo runs, but I'd never go so far as to run them on a track. The guys I run with twice a week do their tempo runs on our normal course, which has hills. I think 10 or more of the guys have run sub 3 hour marathons, so it doesn't seem to be holding them back.

As for pace, I try to run them at my 10k pace. When I'm going up a hill in a 10k, I run a different speed then when I'm going down a hill, right?

"Tempo" runs are designed to improve the body's ability to buffer lactic acid. That requires pushing your effort level so that you produce more lactic acid than your body currently handles over significant time. Thus the 15-20 minute suggestion. Somewhere 4 miles became my magic number as I "hate" workouts based on time. I just perform better with distance goals.

Anyway, depending on level of skill, lactic threshold may be anywhere from a 5k pace (beginners) to half-marathon pace for the pros. Lactic acid causes the burn. Very experienced runners may buffer so well as to not even feel the burn but can feel a sudden fatigue overtake them when they go to far beyond threshold level.

This link might help you get started if you've got some recent race times. http://www.runnersworld.com/cda/trainingcalculator

Finally, do the drills on the road. Learn to deal with pacing hills because you must pace hills when you race too.

xc800runner's description of tempo vs. fartlek is exactly right, as well as jayhawk's descriptor of tempo runs being designed to improve lactic acid buffering. The two are distinctly different, and both serve different purposes.