The Approach-Avoidance Task (AAT) is used to measure and modify approach and avoidance (AA) tendencies which are important in various mental disorders. To date, 46 participants were either instructed to move stimuli in the AAT depending on the content/valence, or depending on a feature that is unrelated to the content and valence. Both ways of instructing participants have fundamental disadvantages that may reduce the validity of the task or the probability of finding existing effects. In this study, both instructions were combined for the first time, with the goal of minimising the downsides of the conventional AATs. Hence, participants had to approach or avoid images depending on both the content of the images (spider, butterfly) and depending on a content-irrelevant feature (portrait vs. landscape format). In an online study, we found that this new dual-feature AAT showed the expected compatibility effect for errors: Responses compatible with AA tendencies were executed more correctly than incompatible responses. Results regarding the RT outcome measures were less clear-cut and depended on data pre-processing decisions. Moreover, the size of the observed compatibility effects was not related to the participants' level of spider fear. In sum, we found a first indication that the dual-feature AAT could be an alternative to previous types of standard-instructions AATs.