Visual perception requires the construction of “neural representations,” patterns of neuronal activity that represent the things we see. The most well-understood component of this construction is the feedforward hierarchical processing pathway, in which neural signals, starting at the eyes, are serially processed to form more integrated and specific representations at successive levels of the visual hierarchy. There is, however, growing evidence that feedback signals from higher to lower levels of the visual hierarchy moderate the quality of neural representations, and may even dictate whether those representations can be perceived.Object-substitution masking (OSM) is a visual “masking” technique used to investigate how neural representations become available to conscious awareness. It is thought to disrupt “reentrant processing,” a form of feedback signaling thought to be required for visual awareness. By disrupting this reentrant processing, OSM is thought to prevent confirmation and conscious awareness of the initial neural representations of briefly presented stimuli. Theories of OSM have traditionally appealed to disruptions of “object-level” neural representations; however, there is evidence for “feature-level” effects in OSM, along with theoretical grounds for predicting that reentrant processing may operate on both object-level and feature-level neural representations. The experiments of this dissertation were performed to test for and investigate feature-level contributions to OSM for the features of tilt, color, and lightness. Stimuli were designed to selectively activate specific types of feature-processing neurons, and parameterized to isolate feature-level contributions to OSM. Experiments 1A-1C established OSM for tilt and color, and showed that masking of these two features can be dissociated. Experiments 2A-2D confirmed this masking using single-feature report. These experiments did not show any effects of similarity between targets and their masking flankers on masking, regardless of whether that similarity was of a task-relevant or a task-irrelevant feature. Finally, Experiments 3A-3C showed masking of targets defined by a single color or luminance feature. These experiments also showed evidence that the color or luminance axis and identity of the target and flankers affected degree of masking, with greater masking of and by some types of targets and flankers than others. This pattern of results is not as would be predicted by previous models of the role of flanker-target similarity in OSM. Instead, they are consistent with a feature-level contribution to OSM that is mediated by neurons that process specific feature signals, such as cone-opponent L/(L+M) and S/(L+M) color signals and (L+M) luminance signals. Furthermore, the efficacy of masking for target color or luminance features aligns with the transmission speeds of low-level feature-processing neurons for those features: Specifically, color or luminance representations processed by faster-signaling neurons are masked more effectively than those processed by slower-signaling neurons. From these patterns of results, a new model of OSM is developed. This new model suggests feature-level contributions to OSM based on relative signal-transmission speeds of low-level feature-processing pathways for those features.