It's a very hard thing to measure, mostly because double-blind studies are unethical, and well-controlled case-control studies are generally difficult. An 85% reduction in head injuries is often cited, but it's based on a study from the mid-'80s that, it turns out, had some pretty substantial methodological flaws, so it's generally agreed that that number overstates helmets' efficacy pretty considerably. More recent studies put the number somewhere around a 20% reduction in head injuries for adults in the US, accompanied by a slight increase in the rate of torsional injuries to the neck because of the larger circumference of a helmeted vs. unhelmeted head. In general, though, helmet use is not an especially good predictor of the outcome of a bike accident; it's much less well correlated with likelihood of death than, say, the cyclists's speed at the time of the accident.