We show that migration flow estimates from population surveys are unreliable: either the flows are not captured in the sample or they are overestimated. This arises from the framing error inherent in this enterprise. The sample frame is the whole population, while the proper frame to estimate migration flows are migrants. A survey designed for the former frame will fail to produce accurate results for the latter. The only exception to this occurs when all flows are large relative to populations as may occur in state-to-state flows. In this case, each flow is large enough relative to the population that the probability that it is not sampled is vanishing small.