A fundamental issue in visual cognition is whether high-level visual areas code objects in a part-based or a view-based (holistic) format. By examining the viewpoint invariance of object recognition, previous behavioral and neuroimaging studies have yielded ambiguous results, supporting both types of representational formats. A critical factor distinguishing the two formats could be the availability of attentional resources, as a number of studies have found greater viewpoint invariance for attended compared to unattended objects. It has therefore been suggested that attention is necessary to enable part-based representations, whereas holistic representations are automatically activated irrespective of attention. In this functional magnetic resonance imaging study we used a multivariate approach to probe the format of object representations in human lateral occipital complex (LOC) and its dependence on attention. We presented human participants with intact and half-split versions of objects that were either attended or unattended. Cross-classifying between intact and split objects, we found that the objectrelated information coded in activation patterns of intact objects is fully preserved in the patterns of split objects and vice versa. Importantly, the generalization between intact and split objects did not depend on attention. Our findings demonstrate that LOC codes objects in a non-holistic format, both in the presence and absence of attention.