Nudix hydrolase 15 (NUDT15) deficiency is strongly associated with thiopurine‐induced myelosuppression. Currently, testing for NUDT15 deficiency is based on the genotyping of the most frequent and clinically characterized no‐function variants, that is, *2, *3 and *9. The Hispanic/Latino‐predominant variant NUDT15 *4 (p.R139H) is classified as “uncertain function” by the Clinical Pharmacogenetics Implementation Consortium, because of insufficient data to ascertain its clinical actionability. In this study, we evaluated the association of NUDT15 *4 with mercaptopurine (6‐MP) tolerance in a retrospective cohort of 1,399 patients with acute lymphoblastic leukemia (ALL) of diverse ancestries. All patients were wildtype for thiopurine methyltransferase gene. Patients were treated with 6‐MP in the context of ALL frontline clinical trials. The tolerated dose of 6‐MP was used to assess drug toxicity during the maintenance phase of ALL therapy. Patients with NUDT15 *1/*4 (n = 16, all of whom self‐identified as Hispanic/Latino) tolerated a significantly lower dose of 6‐MP than did those with NUDT15 *1/*1: median [interquartile range] of 39.0 [21.2–52.8] mg/m2, vs. 62.2 [47.9–71.6] mg/m2, P value < 0.001. No patient homozygous for *4 was detected. In a separate retrospective validation study, six patients were identified as having NUDT15 *1/*4 by routine clinical pharmacogenetics testing and tolerated a 6‐MP median dose of 38.7 mg/m2 (IQR, 33.7–54.0), confirming the need for dose reduction attributed to the NUDT15 *4 variant. These results point to the need to include NUDT15 *4 in pharmacogenetics‐guided thiopurine dosing algorithms.