Estimation of long-term effective population size (N_e) from polymorphism data alone requires an independent knowledge of mutation rate. Microsatellites provide the opportunity to estimate N_e because their high mutation rate can be estimated from observed mutations. We used this property to estimate N_e in allotetraploid wheat Triticum turgidum at four stages of its history since its domestication. We estimated the mutation rate of 30 microsatellite loci. Allele-specific mutation rates μ were predicted from the number of repeats of the alleles. Effective population sizes were calculated from the diversity parameter θ = 4N_eμ. We demonstrated from simulations that the unbiased estimator of θ based on Nei's heterozygosity is the most appropriate for estimating N_e because of a small variance and a relative robustness to variations in the mutation model compared to other estimators. We found a N_e of 32,500 individuals with a 95% confidence interval of [20,739; 45,991] in the wild ancestor of wheat, 12,000 ([5790; 19,300]) in the domesticated form, 6000 ([2831; 9556]) in landraces, and 1300 ([689; 2031]) in recent improved varieties. This decrease illustrates the successive bottlenecks in durum wheat. No selective effect was detected on our loci, despite a complete loss of polymorphism for two of them.