Iconic Memory is not a Feature Soup

Visual sensory memory (also known as iconic memory) is typically characterized as a pre-attentive store, and attention is thought to move information from this rapidly decaying store into the more durable working memory system. If attention is the mechanism that binds visual features into coherent objects, as is assumed by prominent theories, then the contents of iconic memory should be comprised of a collection of visual features rather than bound objects. We recently found that iconic memory capacity for faces is no greater than that of working memory, which could reflect a failure to build holistic faces in the pre-attentive iconic store. In the following two experiments, recall performance was tested at retention intervals that measure iconic and working memory to directly determine whether errors in feature binding were more prevalent in iconic than working memory. In Experiment 1 participants reported the color and orientation of a cued line, and in Experiment 2 they reported the color and shape of a cued object. Standard analyses were applied to the joint distribution of color and orientation (or color and shape) to measure the rate of binding errors (e.g., accurately recalling the color of an object without any knowledge of its shape). In the working memory condition, results replicated previous findings that binding failures are somewhat common. Critically, binding failures were not more common in iconic memory than in working memory, as would be predicted if feature binding only occurred during an attentionally driven transfer into visual working memory. Therefore, we believe that iconic memory is not merely a soup of features that decay independently, and the binding failures observed in working memory are probably a limitation in perception rather than working memory storage.