One of the more valuable contributions of Peter Rossi's article is to remind us of the critical role that shelters play in defining and responding to the problem of homeless families. As Rossi points out, shelters help form our conceptualization of the problem-what kinds of families are homeless and why as well as their number. Perhaps more important, increasingly sophisticated shelters have come to define our policy response to family homelessness. Rossi reports that the number of family shelters soared throughout the middle and late 1980s and that they changed from simply providing a roof and food in an emergency to also providing social, medical, and psychological services for longer periods in more private quarters. And the future promises much more of the same. If the Department of Housing and Urban Development (HUD) follows the suggestions its Assistant Secretary for Community Planning made for New York City, the Clinton homeless family policy will likely center around shelters (New York Commission on the Homeless, 1992). Here we want to extend Rossi's emphasis on the role of shelters in dealing with family homelessness. We make two arguments. One, shelters define the problem of family homelessness and therefore a particular conception of that problem. Two, shelters act to select from the population of poor families those who are worst off in some ways (housing, income, managing their lives, drug or alcohol addictions) and least able or willing to cope with circumstances other poor families do handle. We then explain that one reason why shelters play this role is because the ratio of "worst off, least able" poor families to the total number of poor families is so small that a device is needed to identify these families. Shelters are such a device. In the final section, we fill a lacuna in Rossi's causal analysis with a few ideas as to why shelter growth exploded in the mid-1980s and we explain our disagreement with Rossi's major policy suggestions. Because shelters house such a small proportion of all poor families and because few families stay sheltered very long, Rossi's recommendations are a very inefficient way to end homelessness. Also, the amounts of money entailed in his proposals are so large they are unlikely to be appropriated; and if lesser amounts are appropriated, they are unlikely, absent targeting, to reach families apt to become homeless.