Importance: Latent hepatic Plasmodium vivax hypnozoites provoke repeated clinical attacks called relapses. Only primaquine phosphate kills hypnozoites, and its therapeutic activity may depend on naturally polymorphic cytochrome P450 2D6 isotype (CYP2D6) activity. Objective: To examine the association of impaired CYP2D6 genotypes and CYP2D6 metabolic phenotypes with therapeutic failure of directly observed high-dose primaquine treatment for P vivax malaria relapse. Design, Setting, and Participants: Nested case-control study of patients who, in July 2014, completed a randomized clinical trial of directly observed primaquine treatment for radical cure of acute P vivax malaria in an area of Indonesia where reinfection during 1 year of posttreatment follow-up was improbable. A total of 177 of 180 patients with P vivax malaria completed the clinical trial of primaquine treatment to prevent relapse; 151 were eligible for recruitment as controls. After screening, 59 potential control individuals (no relapse) and 26 potential case patients (relapse) were considered, and 36 controls and 21 cases were enrolled. Exposures: Cases and controls were exposed to P vivax malaria and primaquine therapy but had variable exposure to the enzymatic activity of CYP2D6, classified as impaired by a genotype-determined qualitative phenotype (poor or intermediate), genotype-determined activity score less than 1.5, or a log of the 24-hour pooled urine dextromethorphan-dextrorphan metabolic ratio greater than -1.0. Main Outcomes and Measures: Unadjusted odds ratios (ORs) of relapse with impaired CYP2D6 metabolism determined by genotype or measured by urinary dextromethorphan-dextrorphan metabolic ratio. Results: Among the 21 cases (mean [SD] age, 30.5 [6.3] years; all male) and 36 controls (mean [SD] age, 29.0 [3.6] years; all male), 6 CYP2D6 alleles (*1, *2, *4, *5, *10, and *41) occurred as 12 distinct genotypes, with model activity scores ranging from 0.0 to 2.0. Among 32 patients with genotypic activity scores of 1.0 or less, 18 had experienced relapse, whereas among the 25 with scores higher than 1.0, 3 had experienced relapse (OR, 9.4; 95% CI, 2.1-57.0; P = .001). When the log of the metabolic ratio of dextromethorphan-dextrorphan was -1.0 or less, only 1 of 18 patients experienced relapse, whereas above that threshold (consistent with low metabolic activity), 20 of 39 patients experienced relapse (OR, 18; 95% CI, 2.2-148.0; P = .007). Conclusions and Relevance: Genotype-determined and directly measured impaired levels of CYP2D6 activity were associated with elevated risk of therapeutic failure. These findings suggest a natural variability in CYP2D6-dependent metabolism of primaquine as a key determinant of therapeutic efficacy against latent P vivax malaria.