Optimal input allocation in agriculture leverages production complementarities. For example, improved seeds are generally more responsive to fertilizer than traditional seeds. Thus, inaccurate beliefs about whether seeds are improved may result in sub-optimal fertilizer application. We document precisely this pattern using data from Ethiopia that allows us to compare farmer beliefs about their maize seeds with genotyping data that identify the true genetics of these seeds. We find that 15 percent of farmers believe incorrectly that they are using improved varieties and use far more fertilizer than farmers who correctly believe that they sowed traditional varieties. Conversely, we find that about 15 percent of farmers believe incorrectly that they are growing traditional material and use far less fertilizer than those farmers who correctly believe that they are growing improved material. We extrapolate from our nationally representative sample to estimate the national-level magnitude of fertilizer misallocation due to incorrect seed beliefs.