This study investigated the main effects of social support measures and their stress-buffering effects in two samples of homeless adults (Ns =249 and 219) obtained in the same large county (surrounding Detroit) at different points in time over an 8-year period (1992-1994 and 2000-2002). The findings suggest that the construct of social support, commonly applied to broad community and student samples, can also be usefully applied to at-risk groups such as the homeless. Although the study's main effects and stress-buffering effects were generally consistent with existing research on social support, the findings point out that expected results do not consistently emerge even when similar sampling and measurement techniques are employed. For example, a measure of perceived support showed consistent main and stress-buffering effects on psychological symptoms (but not other outcomes) and social network measures showed less consistent effects on outcomes (e.g., alcohol and drug abuse symptoms). (Contains 3 tables, 5 footnotes, and 3 figures.)