Researchers investigating the link between trade and peace often face a severe problem of list-wise deletion from missing trade data. Attempts to mitigate this problem include assuming that most observations are zero or imputing the values of such flows. We compare two frequently used trade data sets (the Gleditsch data set and the Correlates of War Project data set). We classify individual observations as observed, constructed or missing. We demonstrate that state attributes are systematically related to different categories of trade data. Using Monte Carlo simulations, we also find that replacing some missing data with estimated values tends to inflate the effects of trade in conflict models, although the effects differ by data set.