Structural estimation of welfare effects from new goods largely relies on demand models that include a logit error and the simulation of a counterfactual that removes the new goods. In markets where consumers have both brand preferences and product-type (option) preferences, nested demand models come into play. However, the logit errors in nested logit models not only lead to an inaccurate prediction of shares in the counterfactual but also make welfare estimates sensitive to the number of nests. To deal with these two problems which give rise to implausibly large welfare estimates, I develop an empirical framework to estimate a pure characteristics demand model that allows for option (nest) choices and eliminates the need to rely on logit errors. I then provide an application using scanner data to quantify the welfare effects of four new products simultaneously introduced to the U.S. shampoo market where consumers have strong preferences over options. Results show that consumer welfare increases the most from the shampoo product that offers a niche option. Compared with models featuring logit errors, my approach reduces welfare overestimation by at least 73%. It also provides a better fit and more accurate welfare estimates than the pure characteristics model ignoring options.