BlueRain's right about the essays - make sure you do them so you can develop stamina!

Most people find GMATPrep the most accurate indicator of any of the practice tests (which makes sense, seeing as it is written by the same people who write the real thing!). It also has a standard deviation, just as the real test does. The standard deviation on the real test is about 30 points. I haven't actually seen a published number for SD on GMATPrep so, depending upon how closely, exactly, it mimics the real thing (which we don't know), expect either the same or a little worse (higher number) for GMATPrep.

In other words, even the real test is not that precise in terms of its scoring ability - the SD is fairly wide. (It's actually a pretty good number for a standardized test - it's just that standardized tests in general have fairly wide error margins, especially a test that is only taken by about 200,000 people a year.)

I agree with everything Stacey says above, with just one clarification:

Stacey Koprince wrote: In other words, even the real test is not that precise in terms of its scoring ability - the SD is fairly wide. (It's actually a pretty good number for a standardized test - it's just that standardized tests in general have fairly wide error margins, especially a test that is only taken by about 200,000 people a year.)

The standard error of GMAT scores is a function of the number of questions on the test, and not of the number of people taking the test. If the test had more questions, the error would be lower. The computer adaptive version of the GMAT was designed to be long enough so that the standard error would be 'acceptable' (that is, similar to the standard error of the paper tests) without being so long as to be unduly taxing for the test-taker.