Studies in empirical aesthetics vary largely in the kind of stimuli used, and yet they often make general claims about the effects of stimulus factors like complexity on aesthetic attributes like beauty, pleasure and interest. Here, we wanted to investigate how stable these effects are when different stimulus categories are intermingled rather than presented in homogeneous blocks. We selected five main stimulus categories, each time with a rather abstract and a more figurative or recognizable subset: fractals, geometric patterns, natural images, art photography, and paintings. All ten stimulus categories had twenty images, evenly distributed across five levels of objectively computed complexity values. The resulting 200 images were either presented in 10 homogeneous blocks or in 10 heterogenous blocks, to two groups of >200 participants each. All participants rated how beautiful they found these images, and different subgroups also rated pleasure and interest, or order and complexity, all on 7-point Likert scales. Results indicated that the aesthetic measures varied with the preselected five levels of complexity in different ways for the ten stimulus categories and the five scales. Moreover, the effects differed significantly between homogeneous and heterogeneous blocks. In general, complex interactions were obtained between all the variables, which strongly indicates that many effects are not only stimulus-dependent but also context-dependent (strongly affected by the other stimuli included in the set). This study suggests that researchers have to be cautious in drawing general conclusions about effects of stimulus variables on aesthetic ratings from experiments with limited and homogeneous sets of stimuli.