The evolution of resistance in response to pesticide selection is expected to be delayed if fitness costs are associated with resistance genes. The estimate of fitness costs usually involves comparing major growth traits of resistant versus susceptible individuals in the absence of pesticide. Ideally, a measure of changes in resistance allele frequency over several generations would allow the best estimate of the overall fitness cost of a resistance gene. In greenhouse conditions, we monitored the dynamics of the evolution of the frequencies of six herbicide-resistant mutations (acetolactate synthase, cellulose synthase, and auxin-induced target genes) in the model species Arabidopsis thaliana in a multigenerational study covering five to seven nonoverlapping generations. The microevolutionary dynamics in experimental populations indicated a mean fitness cost of 38%, 73%, and 94% for the ixr1-2, axr1-3, and axr2-1 resistances, respectively; no fitness cost for the csr1-1, and ixr2-1 resistances; and a transient advantage for the aux1-7 resistance. The result for the csr1-1 resistance contrasts with a cost of 37% based on total seed number in a previous study, demonstrating that single generation studies could have limitation for detecting cost. A positive frequency dependence for the fitness cost was also detected for the ixr1-2 resistance. The results are discussed in relation to the maintenance of polymorphism at resistance loci.