I have read that you should stress test your CPU for 24 hours to ensure stability. Does that mean 24 hours straight, or can you do smaller blocks of time?

EG. 6 hour stress test x 4 = 24 hours.

IMO, you can use less time. You will reach max temp in about 15min or less. I have read somewhere that there is a concern that needlessly running Prime95 et al. on a highly OCed CPU for extended periods of time will needlessly increase electromigration. I don't know if this is true but it certainly makes sense. If you have a stock clocked CPU I don't think it makes much difference. If you can run prime small FFTs for an hour and Prime95 blend for an hour, then your machine is stable enough. This is my opinion only, I'm sure many do not agree.

24 hours straight.

I agree. I've always thought the amount of time to stress test a chip is the same as the maximum amount of time you'll use the computer for under stressful conditions. So for me, I usually do a 3-4 hour stress test, since i never use my computer for longer than that at any one time.

From my experience I OC'd and ran Prime95 for 8 hours (blend) and no error
One hour more it stoped and reported an error
I pushed the voltage one notch and ran it again 24 houes with no error
So I consider my rig is stable
I've even read some reviews of overclockers who got errors after 24 hours and had to run 48 hours
Definitely I recommrnd 24 hours continuousely
By time test is getting harder and parts are more stressed
look at results
Self-test 8K passed! At beginning
Self-test 384K passed! After 3 hours
Self-test 1536K passed! After 7 hours
Self-test 4096K passed! After 10 hours

You should do small FFT for 8hrs+. And Blend for 6hrs+.

I run 24 hours at initial stock speed and later at the final OC. I've had failures at 14 and 18 hours.

I don't think that there's any hard and fast rule.