Families of 52 first-admission patients diagnosed with a severe psychiatric disorder were videotaped interacting with the patient. Behavioral coding was used to derive several indices of interaction: base rates of positive and negative behavior by patients and relatives, cumulative affect of patients and relatives (the difference between the rates of positive and negative behaviors), and classification of families as affect-regulated or unregulated. Family-affect regulation reflects positive cumulative affect by both people in a given interaction. Six months after hospital discharge patients were assessed on occurrence of relapse, global functioning, severity of psychiatric symptoms, and quality of life. Relative to affect-unregulated family interaction, affect-regulated interaction predicted significantly fewer relapses, better global functioning, fewer positive and negative psychiatric symptoms, and higher patient quality of life. Most of the predictions by family-affect regulation were independent of severity of psychiatric symptoms at hospital admission, premorbid adjustment, or psychiatric diagnosis. Base rates and cumulative affect of relatives and patients were unrelated to patient relapse, but relative behaviors and cumulative affect did predict other aspects of patient outcome.