This article compares two approaches to housing chronically homeless individuals with psychiatric disabilities and often substance abuse. The experimental Housing First programme offered immediate access to independent housing without requiring psychiatric treatment or sobriety; the control Continuum of Care programmes made treatment and sobriety prerequisites for housing. A total of 225 participants were interviewed prior to random assignment and every 6-months thereafter for 2 years. Data were analysed using repeated measures analysis of variance. Participants randomly assigned to the experimental condition spent significantly less time homeless and in psychiatric hospitals, and incurred fewer costs than controls. A sub-sample recruited from psychiatric hospitals (n = 68) spent less time homeless and more time hospitalized, and incurred more costs than a sub-sample (n = 157) recruited from the streets. Recruitment source by programme interactions showed that the experimental programme had greater effects on reducing hospitalization for the hospital sub-sample and reducing homelessness for the street sub-sample. Three-way interactions including time indicated that in the experimental group, hospitalization and homelessness declined faster for the hospital and street sub-samples, respectively, than for comparable controls. Overall results support the Housing First approach. Copyright © 2003 John Wiley & Sons, Ltd.