This paper proposes a parallel technique for quickly recognizing many different objects. The proposal is based on five approaches for recognition: use a set of two-dimensional characteristic views, describe objects in terms of primitive parts, use simple local properties, use relations defined in terms of relative values of properties, and specify classes using simple constraints on property values. These approaches are not directly supported in the paper with references to relevant psychological research but are presented as reasonable properties of human vision. Based on the approaches, Rosenfeld gives a proposal for a three-stage technique for a parallel system: input the image, find primitive parts, and compute property values; broadcast these properties to object processors; and in each object processor, compare the input description with its stored description, in order to recognize the input. This is simply a proposal designed to encourage researchers to think about parallel techniques and to develop techniques that may fit in the proposed framework.