For most of the world's population, BMI is a fairly reliable estimate of body fat. That being said, BMI is not a direct measurement of fatness. It's a mathematical model that closely correlates to body fatness for most people.
But there are outliers that don't fit within the model. I'm one of them.
For my 70 inches of heigh, according to BMI tables if I were to have a BMI of 18.5 to 24.9, which is a normal healthy adult BMI, I'd weigh between 132 and 174 lbs.
At the moment I'm 250 pounds. According to three seperate and relatively high end devices estimating my body fat percentage using Bio-Impedance (BIA) my body fat percentage is measured to be between 25 and 26%.
Assuming a worst case of 26% body fat at my current weight, I am carrying around 65 pounds of fat. That would make my lean body mass 185 pounds.
But, if my lean body mass is 185 pounds, how can my ideal weight top out on the BMI chart at 174? The results of these two seperate measurement systems are incongruent. Either a) all three BIA analyses were grossly incorrect or b) I am some sort of outlyer to the BMI model.
Given the likelyhood of a and b, I have to make the assumption that b is true and that BMI doesn't apply to my body type. This isn't exceedingly surprising. I'm a big guy. With only casual weight training, I can bench press about 300 pounds, which is something many serious body builders strive for. BMI doesn't account for body composition completely.
So then, should BMI be gotten rid of as a measurement of fatness? Probably not, but I think it may be time to reconsider how BMI is used.
For large groups of people it is an excellent tool for measuring fatness, as it is for most of the population accurate, at least according to the CDC. For an individual, trying to measure his or her individual fitness there are better tools readily available.
For an individual, estimating body fat by BIA is likely a superior method. The BIA method is at least very strongly correlated to actual body fat, and in most studys I've read is shown to agree within 5% with more accurate methods like DXA analysis. Additionally, it should be a much more inclusive method, which fits large people as well as smaller people (studies suggest that for some asian people obesity begins at BMI 26 instead of 30 for caucasians) better than BMI can.