I have tested a lot of different voltage references and put together a chart detailing their specifications along with their output voltage as measured on my 34401A.
I tried to list the specification limit, not the "typical" deviation figure. I didn't want to compare one model's typical as compared to another's maximum. Occasionally I had to make an educated guess as to its maximum value. Case-in-point: the beloved LTZ1000. Linear likes to play up its 0.05ppm/C temperature drift specification without defining it. It appears to be a typical value. I guessed that its maximum might be 0.2ppm/C (although it could be 1ppm/C like some other models). The LTZ1000CH is the only voltage reference detailed on my attachment which I didn't have on hand to measure its actual output voltage.
I colored coded my opinion of the specification. For example, The LTZ1000 and LM399 have red accuracy specifications, but bright green temperature and long term stability ratings. I tried to be fair in those ratings and based them on their relative figures compared to the others listed.
The LM4128AMF-3.0 which I tested had a high noise level and poor voltage stability. It was by far the worst one tested in those regards.
I included a temperature compensated 1N4573A-1 zener in my tests for comparison. It had good voltage stability and a noise level comparable to the average voltage reference. The economy BZX79-B6V8,143 zener that I also tested for comparison had both a high noise level and poor voltage stability.
Most of the voltage references tested were purchased either from Mouser, Digi-Key or Arrow. The two AD584's tested were inside the eBay Chinese boxes which are available. The AD688BQ was recently obtained from
http://www.ebay.com/itm/182537476611 for $18.91 + $4.00 shipping. It meets its 0.02% voltage accuracy specification for its +/-10V output. I'll probably add some trimmers to improved that accuracy further. Its voltage stability has been good in its initial tests. It's nice not to have to play games to alter the voltage to obtain 10 volts. Having a similarly accurate 20 volts output can double the precision of tests of some equipment.
Only testing one of most voltage references makes my test of them be not that meaningful. The LM399AH was the only one where I tested two. I wish that the tested ISL21009BFB's wouldn't have been discontinued. The specs of their replacements aren't as good.
Hopefully the attachment is useful for some. I tried to provide a fair comparison of the different models. Hopefully I didn't let any typical values slip through and that my approximations of a few maximum specifications aren't too far off the manufacturer's limitations. Several additional models were just added 6July2017 and 6August2017. At the bottom of the attached chart I just added the average measured accuracy of all of the voltage references shown which had a specification sheet accuracy of better than 0.10%. That average accuracy of 0.000486% might be one way of verifying that my Agilent 34401A meter which was used meets its 0.0015% specification sheet accuracy. 35 different voltage references were used in calculating that average.