Fuzzy rule systems are an important element in the arsenal of automated control, as they are able to process the input provided by sensors and provide their output in very small times. Lately, there has been augmented interest in utilizing fuzzy rule sys-tems in a wider range of applications, due to the in-tuitive way in which they represent and utilize knowledge. In these contexts, input is not always readily available from a sensor but often provided as the questionable output of another expert system or even totally missing. In this paper we propose a novel approach to rule evaluation that is able to operate under such uncertain conditions and evaluate it by applying it to the case of facial expression analysis. Our approach has a possibilistic, rather that prob-abilistic, flavor.