Proxy Means Testing (PMT) is a popular method to target the poor in developing countries. PMT usually relies on survey-based consumption data and assumes they are measured with random errors – an assumption that has been challenged by recent literature. This paper brings causal evidence on the impact of non-random errors on PMT performances. Using a survey experiment conducted in Tanzania in which eight alternative consumption questionnaires were randomly distributed across households, I compare the performances of PMT relying on error-prone consumption data with those of a PMT using gold standard consumption data. Results show that non-random errors reduce the predictive performances of PMT by a magnitude ranging from 5 to 27 percent, which in turn induce a 10 to 34 percent increase in the incidence of targeting errors (using the typical PPP $1.25 poverty line). Impacts on the relative distribution of households are nonetheless small and non-significant.