I would take exception to the statement that ANSYS seems to have the most widespread use. That is very much a function of where one resides, both geographically and from an industry standpoint. Although I do know of areas where ANSYS is a major player, it is a virtual unknown in other areas (North American Automotive, for instance).
Don't get me wrong, ANSYS is a decent code.
In deciding what code to utilize there are several factors:
1) what types of analysis are you doing:
In the FEA field, largely you get what you pay for. The lower-end codes are able to do less. Any applied mechanics grad student can make a limited-purpose FEA code, but this code, though inexpensive, may not have the capability of doing many things that you want it to do. There are three levels of complexity: material nonlinearity, geometric nonlinearity, and boundary (contact) nonlinearity. Different codes can do some of these. Some of the less expensive codes can't do any nonlinearity (although one may not need it). The more expensive, multi-purpose codes can do all of these, in many neat ways, but at a price (basically, you largely get what you pay for).
Your locked-in price of 4,000 pounds (@$6,000) will not buy you one of the better packages. Note: I am assuming that your price is on a yearly basis. A $6,000/year budget will buy you a decent FEA solver, but not one of the higher-end solvers.
2) What types of computers will you be running this on.
Some codes are not ported to some platforms. More importantly, some of the analysis you may want to do may require more computing horsepower than you are equipped with. Alternately, some of these codes are ported to Windows NT machines, which are generally somewhat cheaper (though less reliable) than equivalent UNIX boxes.
3) As dsi noted, what CAD/CAE programs do you already use.
Some FEA manufacturers are tightly integrated with particular CAD packages, so the existence of a particular cad package at you site may help you in going from cad to cae.
4) What level of experience do you and the other engineers have with FEA.
If you are all novice users, you may not be prepared to do some of the more elaborate analyses, so the additional cost may not be useful for you. On the other hand, if you have capable people with other FEA experience, you may quickly find that the higher-end packages allow you to do things which you had previously not even considered. A higher end package (10,000 pounds/$15,000) may end up allowing you to do analyses which save you more than the incremental cost.
5) Finally, some of the FEA solvers do not have their own pre/post processors, so there may be an additional cost in buying third-party software to build your models.
In short, there is not an easy answer to your question, but here are the additional questions that can help you in your decision.
I would recommend talking to various FEA software distributors, and throwing these types of questions to them. If you know somebody who has done FEA (a friend, former classmate), pick their brains for awhile, and then invite software distributors to give you their sales pitch. A final word of advice: Pay attention to the BS factor from the software distributors, and weight it heavily. Try to push them and see if they'll try and promise the world in terms of capabilities. It has been my experience that the better FEA software companies tend to give caveats, because they would rather slowly build a very strong and loyal group of users at a slow pace. They don't want to disappoint anybody by promising too much too soon (although they will certainly acknowledge that given experience many things can be done). The less reputable software companies give a pitch like "Any CAD operator can perform FEA with our code". This is bunk--no code is able to allow an unqualified person to generate reasonable results consistently. You still need trained engineers to run this software.
If you ignore everything else I have said, keep this last statement in mind.
Good luck and happy shopping.
