Recently on Edmunds.com, ratings at http://www.informedforlife.org/ were pointed to as “the most comprehensive evaluation of safety in accidents to date.”
They are comprehensive in some sense but definitely not the whole story. Any safety ratings which do not consider braking and handling–that is, active safety, or the vehicle’s ability to avoid an accident in the first place–don’t tell the whole story.
What you want to know is, if I drive this car, how likely is it that I’m injured? Crash test ratings tell you something about the probability of injury given a crash, but to get the probability of injury, you also need to know the odds of a crash in the first place. Obviously the driver is a critical component of this, but the vehicle matters a great deal, too. An alert driver has a far better chance of avoiding an accident in a good-handling and fast-braking car than in some lumbering hulk. The provided ratings don’t take that into account. For example, there’s no way a cow like the Uplander is actually safer than the incredibly nimble RX-8.
(On a side note, those “comprehensive” ratings include weighting for the crash results for rear passengers. If you never have rear passengers, why on earth should that get any weight in the final rating? I care because I have kids in the back, but if I didn’t, why should that factor in?)