A number of studies have demonstrated sound-symbolism effects in adults and in children. Moreover, recently, ERP studies have shown that the sensitivity to sound-symbolic label–object associations occurs within 200 ms of object presentation (Kovic et al., 2010). It was argued that this effect may reflect a more general process of auditory–visual feature integration where properties of auditory stimuli facilitate a mapping to specific visual features. Here we demonstrate that the sound-symbolism effect is design dependent, namely — it occurs only when mapping from auditory to visual stimuli and not vice verse. Two groups of participants were recruited for solving the categorization task. They were presented them with 12 visual stimuli, half of which were rounded and another half of angular shapes. One group was trained to classify the rounded objects as ‘takete’ and the rounded ones as ‘maluma’, whereas the other group mapped ‘takete’ to rounded and ‘maluma’ to angular shapes. Moreover, half of these two groups heard the label before seeing the objects, whereas the other half was given the label after perceiving the object. The results revealed the sound-symbolism effect only in the group which was trained on the auditory–visual objects mapping and not in the one trained on the visual–auditory mappings. Thus, despite the previous findings we demonstrate that the sound-symbolism effect is not significant per se, but design-dependent and we argue that the sound brings up a mental image that is more constrained than the sounds brought up by a picture.