The brain solves visual object recognition through a cascade of feedforward computations that culminate in a powerful representation in the inferior temporal (IT) cortex. This process enables rapid recognition of objects despite variations in appearance. The algorithm underlying this solution remains poorly understood, but recent evidence suggests that understanding it requires analyzing neuronal and psychophysical data to sift through computational models based on small, canonical subnetworks. Core object recognition is the ability to rapidly identify objects without pre-cuing, even under identity-preserving transformations. This task is challenging due to the vast variability in object appearances and the need for invariance across different viewing conditions. The visual system must distinguish between similar objects and recognize them despite changes in position, scale, lighting, and clutter. Neuronal evidence suggests that the ventral visual stream, including the IT cortex, plays a critical role in this process. The IT cortex houses key circuits for object recognition, with populations of neurons providing a robust representation of object identity. This representation is supported by a combination of feedforward and feedback mechanisms, allowing the brain to efficiently process and recognize objects. The IT population representation is characterized by a spatiotemporal pattern of spikes that enables the brain to decode object identity. This representation is robust to variations in object appearance and is supported by a hierarchical organization of visual areas. The ventral stream processes visual information in a series of stages, with each stage contributing to the gradual untangling of object identity manifolds. The algorithm that produces the IT population representation likely involves a combination of feedforward and feedback mechanisms, with each stage of the ventral stream contributing to the overall processing of visual information. This process allows the brain to rapidly and accurately recognize objects, even under varying conditions. Understanding this algorithm requires integrating data from multiple levels of analysis, including neuronal activity, population responses, and computational models.The brain solves visual object recognition through a cascade of feedforward computations that culminate in a powerful representation in the inferior temporal (IT) cortex. This process enables rapid recognition of objects despite variations in appearance. The algorithm underlying this solution remains poorly understood, but recent evidence suggests that understanding it requires analyzing neuronal and psychophysical data to sift through computational models based on small, canonical subnetworks. Core object recognition is the ability to rapidly identify objects without pre-cuing, even under identity-preserving transformations. This task is challenging due to the vast variability in object appearances and the need for invariance across different viewing conditions. The visual system must distinguish between similar objects and recognize them despite changes in position, scale, lighting, and clutter. Neuronal evidence suggests that the ventral visual stream, including the IT cortex, plays a critical role in this process. The IT cortex houses key circuits for object recognition, with populations of neurons providing a robust representation of object identity. This representation is supported by a combination of feedforward and feedback mechanisms, allowing the brain to efficiently process and recognize objects. The IT population representation is characterized by a spatiotemporal pattern of spikes that enables the brain to decode object identity. This representation is robust to variations in object appearance and is supported by a hierarchical organization of visual areas. The ventral stream processes visual information in a series of stages, with each stage contributing to the gradual untangling of object identity manifolds. The algorithm that produces the IT population representation likely involves a combination of feedforward and feedback mechanisms, with each stage of the ventral stream contributing to the overall processing of visual information. This process allows the brain to rapidly and accurately recognize objects, even under varying conditions. Understanding this algorithm requires integrating data from multiple levels of analysis, including neuronal activity, population responses, and computational models.