High expressed emotion in relatives has been shown to predict a poorer outcome in a range of disorders and cultures. However, variation in predictive power is considerable, and this could partly be due to low inter-rater reliability. Sixty-nine relatives of 47 acutely admitted schizophrenic patients were interviewed using the Camberwell Family Interview. Audiotapes were assessed by pairs of raters drawn from a group of seven approved raters. Inter-rater reliability was good for criticism, hostility, emotional over-involvement (as a six level scale) and expressed emotion index (EOI cut-off score 2-3); and it was fair, but unsatisfactory, for emotional over-involvement (binary scales), warmth, positive remarks and expressed emotion (EOI cut-off score 3-4). These findings suggest that the warmth scale and the dichotomized versions of the EOI scale should be used with caution. The influence of warmth on the outcome in schizophrenia may have been underestimated because of low reliability.