The artemisinin-based combination therapies artemether-lumefantrine (AL) and amodiaquine (AQ) plus artesunate have been adopted for treatment of Plasmodium falciparum malaria in many African countries. Molecular markers of parasite resistance suitable for surveillance have not been established for any of the component drugs in either of these combinations. We assessed P. falciparum mdr1 (Pfmdr1) alleles present in 300 Tanzanian children presenting with uncomplicated falciparum malaria, who were enrolled in a clinical trial of antimalarial therapy. Pfmdr1 genotype analysis was also performed with isolates from 182 children who failed AQ monotherapy and 54 children who failed AL treatment. Pfmdr1 alleles 86Y, 184Y, and 1246Y were more common among treatment failures in the AQ group than among pretreatment infections. The converse was found in the AL-treated group. Children presenting with the 86Y/184Y/1246Y Pfmdr1 haplotype and treated with AQ were significantly more likely to retain this haplotype if they were parasite positive during posttreatment follow-up than were children treated with AL (odds ratio, 33.25; 95% confidence interval, 4.17 to 1441; P, <0.001). We conclude that AL and AQ exert opposite within-host selective effects on the Pfmdr1 gene of P. falciparum.