We analyzed data for San Francisco intravenous drug users entering treatment, April 1986-September 1988 (N = 7,660). The proportion of cases reporting any needle sharing in the month preceding treatment decreased from 50 percent in 1986 to 28 percent in 1988. Similar decreases were reported by two longitudinal cohorts (needle sharing by the same individuals) admitted in 1986 and 1987 (n = 303), and in 1986 and 1988 (n = 205). In a multiple logistic regression model four variables predicted needle sharing: earlier time of admission, cocaine use, younger age, and being White rather than Black.