R.A. Fisher and H.J. Muller argued in the 1930s that a major evolutionary advantage of recombination is that it allows favorable mutations to be combined within an individual even when they first appear in different individuals. This effect is evaluated in a two-locus, two-allele model by calculating the average waiting time until a new genotypic combination first appears in a haploid population. Three approximations are developed and compared with Monte Carlo simulations of the Wright-Fisher process of random genetic drift in a finite population. First, a crude method, based on the deterministic accumulation of single mutants, produces a waiting time of 1/square root of N mu(2) with no recombination and [formula: see text] with recombination between the two loci, where mu is the mutation rate, N is the haploid population size, and R is the recombination rate. Second, the waiting time is calculated as the expected value of a heterogeneous geometric distribution obtained from a branching process approximation. This gives accurate estimates for small values of N mu large. The estimates for small values of N mu are considerably lower than the simulated values. Finally, diffusion analysis of the Wright-Fisher process provides accurate estimates for N mu small, and the time scales of the diffusion process show a difference between R = 0 and for R >> 0 of the same order of magnitude as seen in the deterministic analysis. In the absence of recombination, accurate approximations to the waiting time are obtained by using the branching process for high N mu and the diffusion approximation for low N mu. For low N mu the waiting time is well approximated by 1/the square root of 8N2 mu(3). With R >> 0, the following dependence on N mu is observed: For N mu > 1 the waiting time is virtually independent of recombination and is well described by the branching process approximation. For N mu approximately equal to 1 the waiting time is well described by a simplified diffusion approximation that assumes symmetry in the frequencies of single mutants. For N mu << 1 the waiting time is well described by the diffusion approximation allowing asymmetry in the frequencies of single mutants. Recombination lowers the waiting time until a new genotypic combination first appears, but the effect is small compared to that of the mutation rate and population size. For large N mu, recombination has a negligible effect, and its effect is strongest for small N mu, in which case the waiting time approaches a fixed fraction of the waiting time for R = 0. Free recombination lowers the waiting time to about 45% of the waiting time for absolute linkage for small N mu. Selection has little effect on the importance of recombination in general.