This paper uses a large administrative dataset providing 105,149 daily observations on 3,176 youth aged between 18 and 24 years using emergency homeless shelters from January 1, 2014, to December 31, 2019, in Calgary, Alberta, Canada. Using k-mean cluster analysis we classify youth who use emergency homeless shelters by their intensity of shelter use. In Calgary, most youth use homeless shelters as a temporary and infrequent refuge when housing is lost. For these youth, shelters are an effective and relatively inexpensive approach for helping them resolve their homelessness. A significant number, however, rely on shelters for much longer stays. For these youth, shelters are ineffective and expensive responses to their homelessness. To understand how youth are changing their use of shelters over time, we introduce time series measuring first admissions and readmissions to shelters. We show that how youth use shelters, and how that use has changed over time, differs for youth who self-identify as Indigenous, Caucasian and visible minority.