I study how the misallocation of new technology to individuals, who have low ex post returns to its use, affects learning and adoption behaviour. I focus on anti-malarial treatment, which is frequently over-prescribed in many low-income country contexts where diagnostic tests are inaccessible. I show that misdiagnosis reduces average therapeutic effectiveness, because only a fraction of adopters actually have malaria, and slows the rate of social learning due to increased noise. I use data on adoption choices, the timing and duration of fever episodes, and individual blood slide confirmations of malarial status from a pilot study for a new malaria therapy in Tanzania to show that individuals whose reference groups experienced fewer misdiagnoses exhibited stronger learning effects and were more likely to adopt.