Work on pattern recognition pushes us into the problem of the definition of a stimulus. One approach has been to assume that a stimulus is best considered as a collection of smaller units or elements commonly termed features. Experimental work centers around the discovery of such units, or their manipulation. A major purpose of this chapter is to question this approach. It is suggested that features do not provide a satisfactory unit of analysis because they lack autonomy where feature theory would expect such to occur. It is further suggested that the reason for this lack of autonomy is that features represent the products of other basic processes. The problem of the definition of a stimulus is better approached as one of differentiation: what controls the manner in which the stimulus is "parsed" or "carved-up" into units; whether these things are called features or objects. From this view features will be "flexible"; they will appear and disappear as conditions change. Experimental analysis should focus on the isolation of the conditions that control differentiation. Thus, the specification of a stimulus becomes the description of the engagement of the processes of differentiation.