Pandemonium is a model of bottom-up pattern recognition first offered by Selfridge. In the model, different "demons" operate to identify letters at different levels. This model is a feature detection model, offered in response to the failure of template matching models to best describe real perceptual and pattern recognition processes. In the model, each level of demon does something different, and largely independent of the others. Image demons basically record the perception of visual information. Feature demons "yell" whenever they "see" a specific stimulus in the proper location/orientation to their specific focus. Cognitive demons listen to the yelling of feature demons, and each cognitive demon assesses whether threshhold "yeling" occurs for enough features for that Cognitive demon to yell out its letter (or presumably, some other kind of stimulus like a letter, object, etc.). The Decision demon listens to competing Cognitive demons to determine what letter is most likely what has been perceived.
Although not everyone agrees that this model best describes actual pattern recognition, it has been used as a model for Hebbian neural assemblies, and some have suggested it could be adapted to accommodate even top-down processes. I am not sure about any of that, but it certainly was an important idea that was made easy to understand through the architecture, and also that it was influential in pushing forward ideas about pattern recognition and perception.
Super-cool website that lets you play with a Pandemonium model is here. Very fun, students love it:
http://epsych.msstate.edu/descriptive/Vision/swamp/Pandemonium/index.html