For a trait influenced by multiple gene loci with approximately equal, additive effect, the trait distribution will approximate a normal distribution (bell curve) over an arbitrary scale of 1-14 units (top row). The probability of survival follows a fitness function that varies over the range of phenotypes [middle row]. As a result, the trait distribution after selection is modified in a predictable manner [bottom row]. The change in trait distribution can be described in terms of the effect on mean and variance.

Under Directional Selection (left), the trait mean is shifted (here, it increases), and the variance is skewed (unbalanced towards the left or right tail (as shown here)) over the course of a single generation; the variance in the offspring generation will be normal around the new mean. Under Stabilizing Selection (middle), both tails are truncated: the trait mean remains the same, but the trait variance is reduced. Under Disruptive Selection, both tails are favored at the expense of the modal phenotypes: the mean remains constant, but the trait variance is increased. [The technical descriptions of the distributions after stabilizing and disruptive selection are leptokurtotic ("pointed") and platykurtotic ("flattened"), respectively']

Selection in each case means that the genetic composition of the next generation is modified. This is most easily appreciated in the case of directional selection, where allele(s) at each locus that increase the trait measure are favored by selection: the frequency of those alleles will be increased in next generation.