The estimation ofdemand models for differentiated products often relies on data from multiplegeographic markets and time periods. The fact that firms do not offer someproducts in some markets generates a problem of endogenous selection in demandestimation. The non-additivity of demand unobservables in firms' expectedprofit and the potential multiplicity of equilibria in the entry game, makethis selection problem challenging. Standard two-step methods to account forselection provide inconsistent estimates of structural parameters. Recentproposed methods require parametric assumptions on the distribution ofunobservables as well as repeatedly computing for equilibria in the game. Inthis paper, we show the identification of demand parameters in a structuralmodel of demand, price competition, and market entry that allows for anonparametric distribution of demand unobservables. We propose a two-stepmethod that extends standard methods to control for selection bias. Weillustrate our method using simulated data and real data from the airlineindustry. We show that not accounting for endogenous product entry generatessubstantial biases that can be even larger than those from ignoring priceendogeneity.