Genetic drift occurs when the genetic state of a population (usually represented in a theoretical model by the frequency of one or more alleles) changes randomly from one generation to the next because of sampling error. (A more detailed discussion is presented in the node genetic drift.) There is some probability that the frequency of adeleterious mutation will increase above the equilibrium level imposed by selection and mutation (the mutation-selection balance) –– a level that is only consistently realised in a population of infinite size; i.e. that does not experience genetic drift (see mutation load). When the drift of deleterious mutations causes a reduction in the mean fitness of a population on average, then there is a drift load.
The drift load can be partitioned into being segregational or fixed. If the population remains polymorphic, such that fitter alleles coexist with the deleterious mutant allele, then the load is segregational. At a particular locus, the reduction of mean fitness is a transitory state; put simply, the frequency of the mutant allele may later drift low enough to restore and even increase the contribution of that locus to mean fitness. On the other hand, if the mutant allele has fixed in the population, then the genetic state of that locus is rendered constant until another mutation arises.
Because of its non-transitory nature, it is the fixed drift load that can cause a cumulative decline in mean fitness. This is often refered to as a mutational meltdown process.