Perceptra is a simple pattern classification algorithm that uses a compositional hierarchy of binary neurons called binons. Each binon represents a class or category. General categories are lower in the hierarchy and more specific combinations of categories are higher in the hierarchy. In this respect, binons are similar to perceptual schemata in schema theory or category prototypes / clusters in implicit category learning. When grounded, they are a possible implementation of Barsalou’s perceptual symbols (Barsalou 1999). For pattern classification the lowest level binons represent features extracted from stimuli. These features are the invariant shape and contrast patterns formed from the ratios between the widths and intensities of the perceived objects. These ratios are calculated by subtracting the logarithms of their values as described in the Weber-Fechner Law. Perceptra starts out with no binons and adds new binons to its network based on the coincidence and novelty of the features and their combinations. Its simplicity allows it to scale well over multiple levels of abstraction. It is sense independent and multimodal. It can recognize patterns of objects independent of their position, width and intensity. It has achieved over an 80% recognition rate on handwritten digits mapped onto a one-dimensional array of sensors.