IMPORTANCE Working memory trainingmay help children with attention and learning difficulties, but robust evidence from population-level randomized controlled clinical trials is lacking. OBJECTIVE To test whether a computerized adaptive working memory intervention program improves long-term academic outcomes of children 6 to 7 years of age with low working memory compared with usual classroom teaching. DESIGN, SETTING, AND PARTICIPANTS Population-based randomized controlled clinical trial of first graders from 44 schools in Melbourne, Australia, who underwent a verbal and visuospatial working memory screening. Children were classified as having low working memory if their scores were below the 15th percentile on either the Backward Digit Recall or Mister X subtest from the AutomatedWorking Memory Assessment, or if their scores were below the 25th percentile on both. These children were randomly assigned by an independent statistician to either an intervention or a control arm using a concealed computerized random number sequence. Researchers were blinded to group assignment at time of screening.We conducted our trial from March 1, 2012, to February 1, 2015; our final analysis was on October 30, 2015.We used intention-to-treat analyses. INTERVENTION Cogmed working memory training, comprising 20 to 25 training sessions of 45 minutes' duration at school. MAIN OUTCOMES AND MEASURES Directly assessed (at 12 and 24 months) academic outcomes (reading, math, and spelling scores as primary outcomes) and working memory (also assessed at 6 months); parent-, teacher-, and child-reported behavioral and social-emotional functioning and quality of life; and intervention costs. RESULTS Of 1723 children screened (mean [SD] age, 6.9 [0.4] years), 226 were randomized to each arm (452 total), with 90% retention at 1 year and 88%retention at 2 years; 90.3%of children in the intervention arm completed at least 20 sessions. Of the 4 short-term and working memory outcomes, 1 outcome (visuospatial short-term memory) benefited the children at 6 months (effect size, 0.43 [95%CI, 0.25-0.62]) and 12 months (effect size, 0.49 [95%CI, 0.28-0.70]), but not at 24 months. There were no benefits to any other outcomes; in fact, the math scores of the children in the intervention arm were worse at 2 years (mean difference, -3.0 [95%CI, -5.4 to -0.7]; P = .01). Intervention costs were A$1035 per child. CONCLUSIONS AND RELEVANCE Working memory screening of children 6 to 7 years of age is feasible, and an adaptive working memory training program may temporarily improve visuospatial short-term memory. Given the loss of classroom time, cost, and lack of lasting benefit, we cannot recommend population-based delivery of Cogmed within a screening paradigm. TRIAL REGISTRATION anzctr.org.au Identifier: ACTRN12610000486022.