An analysis of existing studies on the expressed emotion (EE) concept shows that the predictive validity of the EE index depends on the definition of relapse. Whilst symptomatic relapses are recorded in patients from high-EE families, about 3 times as often as in patients from low-EE families, a comparison of hospitalization rates reveals a difference of only approximately 1.5 times. In the Münster Families Study, a prospective intervention study to evaluate the therapeutic impact of relatives' groups, the predictive value of the familial emotional atmosphere could be measured in a prospective design. Ninety-nine DSMIII Schizophrenics, 72 male, 27 female, mean age 27 (SD 6.6), mean duration of illness 5.5 (SD 4.3) were followed up. In the first eighteen months patients from high-EE- families had a risk of rehospitalisation approximately 1.5 (p < 0.05). Further analysis showed that relapse rates among patients from high-EE and low-EE families were roughly the same for patients with a duration of illness of less than 4.5 years but were significantly different after a longer duration (p = 0.019). Besides a high EE rating, resignation of relatives was also a significant (p = 0.000) predictor of rehospitalisation for high and — even more — low EE families. Possible causes of these findings and therapeutic consequences are discussed.