Evaluative conditioning refers to the observation that the mere paired presentation of a neutral stimulus (CS) with a liked or disliked stimulus (US) may result in the neutral stimulus itself acquiring positive or negative valence. In most studies, the CS is an autonomous, invariant stimulus, and the subject directly experiences both CS and US. In this experiment, we investigated whether evaluative conditioning can be extended to a situation wherein the CS is no more than an invariant element of a complex, variable stimulus configuration, and wherein the subject experiences the CS–US co-occurrences indirectly, i.e. by observing a socius who is exposed to the CS–US pairings and facially expresses either liking or disliking the US. During acquisition, subjects watched video-taped sequences of an actor drinking a glass containing a liquid and facially expressing either liking or disliking the drink. The stimulus element which was systematically paired with the actor's facial expression of liking or disliking, was whether the glass contained a ‘foot’ or no ‘foot’ (CS), while other characteristics of the scenes were systematically varied and paired equally often with an expression of like and dislike. Next, valence ratings were obtained for pictures in which the CS element (foot/no foot) was embedded. A clear observational evaluative learning effect could be demonstrated when the feature CS was embedded in objects identical to those presented during learning, but not when it was embedded in new objects. These data demonstrate the possibility of vicarious evaluative conditioning of an embedded stimulus element, but probably at a lower level of abstraction than intended.