Best option is to omit SPSS and start using R (http://www.r-project.org/) with RStudio interface (http://www.rstudio.com/). There is no reason for professional scientist use to software like SPSS.
I agree with above comments, it depends on how you are going to use the software...
For simple frequency analyses and comparisons, any version will do..
If you want to avoid the costs, (as Rabie has mentioned) an open source software called PSPP (www.gnu.org/s/pspp/) provides a free alternative to SPSS..
In addition, the CDC software Epi Info (wwwn.cdc.gov/epiinfo/) is also free and can provide a more user friendly data entry interface..
Well, of course the latest version (22) is the most advanced. However there are always some drawbacks like bugs within the first half year of a version and fixes are neccessary. All versions up to 21 run totally fine and within the next month the final fixes to version 22 should be published so I would hardly recommend getting 22.
However functionality is not only an issue of the version number but mostly of the package. The most basic SPSS you can get is "SPSS Statistics Base". There are "higher" and more expensive packages with way more functionality. Your best bet would be figuring out for yourself what exactly you need and then asking an IBM Consultant which is the best package for your needs.
For Windows there is an other problem: Windows 8. On some machines it just won't work. So SPSS is best used on Windows 7 or Mac OS.
However:
Using R would be an alternative but when you are looking for "Something like SPSS" it's not a choice. PSPP - well, I keep away from that. I tried it but it was just a pain.
Most the recent versions should do just fine. Of course just like any other software, there, are always hick-ups. I have used various versions prior to and including 21, and all seems to run fine. I am not sure, about version 22, but as suggested by the other colleagues it should do just fine. Good luck.